]> git.cameronkatri.com Git - mandoc.git/blobdiff - mandocdb.c
git.ckatri.com / mandoc.git / blobdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | inline | side by side
When checking the validity of cross references with -Tlint,
[mandoc.git] / mandocdb.c
diff --git a/mandocdb.c b/mandocdb.c
index 824a72d734f1c7293eba594c42b2b87a3e38457d..2a8ab7a0370948006c4b580c58b0e318014caced 100644 (file)
--- a/mandocdb.c
+++ b/mandocdb.c
@@ -1,64 +1,63 @@
-/*     $Id: mandocdb.c,v 1.50 2012/06/08 10:43:01 kristaps Exp $ */
+/*     $Id: mandocdb.c,v 1.251 2017/06/24 14:38:32 schwarze Exp $ */
 /*
  * Copyright (c) 2011, 2012 Kristaps Dzonsons <kristaps@bsd.lv>
 /*
  * Copyright (c) 2011, 2012 Kristaps Dzonsons <kristaps@bsd.lv>
- * Copyright (c) 2011, 2012 Ingo Schwarze <schwarze@openbsd.org>
+ * Copyright (c) 2011-2017 Ingo Schwarze <schwarze@openbsd.org>
+ * Copyright (c) 2016 Ed Maste <emaste@freebsd.org>
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
  * copyright notice and this permission notice appear in all copies.
  *
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
  * copyright notice and this permission notice appear in all copies.
  *
- * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHORS DISCLAIM ALL WARRANTIES
  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR
  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  */
  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  */
-#ifdef HAVE_CONFIG_H
 #include "config.h"
 #include "config.h"
-#endif
 
 
-#include <sys/param.h>
+#include <sys/types.h>
 #include <sys/stat.h>
 #include <sys/stat.h>
+#include <sys/wait.h>
 
 #include <assert.h>
 #include <ctype.h>
 
 #include <assert.h>
 #include <ctype.h>
+#if HAVE_ERR
+#include <err.h>
+#endif
 #include <errno.h>
 #include <fcntl.h>
 #include <errno.h>
 #include <fcntl.h>
+#if HAVE_FTS
 #include <fts.h>
 #include <fts.h>
-#include <getopt.h>
+#else
+#include "compat_fts.h"
+#endif
+#include <limits.h>
+#if HAVE_SANDBOX_INIT
+#include <sandbox.h>
+#endif
+#include <stdarg.h>
 #include <stddef.h>
 #include <stddef.h>
+#include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 
-#include <ohash.h>
-#include <sqlite3.h>
-
+#include "mandoc_aux.h"
+#include "mandoc_ohash.h"
+#include "mandoc.h"
+#include "roff.h"
 #include "mdoc.h"
 #include "man.h"
 #include "mdoc.h"
 #include "man.h"
-#include "mandoc.h"
-#include "mandocdb.h"
-#include "manpath.h"
-
-/* Post a warning to stderr. */
-#define WARNING(_f, _b, _fmt, _args...) \
-       do if (warnings) { \
-               fprintf(stderr, "%s: ", (_b)); \
-               fprintf(stderr, (_fmt), ##_args); \
-               if ('\0' != *(_f)) \
-                       fprintf(stderr, ": %s", (_f)); \
-               fprintf(stderr, "\n"); \
-       } while (/* CONSTCOND */ 0)
-/* Post a "verbose" message to stderr. */
-#define        DEBUG(_f, _b, _fmt, _args...) \
-       do if (verb) { \
-               fprintf(stderr, "%s: ", (_b)); \
-               fprintf(stderr, (_fmt), ##_args); \
-               fprintf(stderr, ": %s\n", (_f)); \
-       } while (/* CONSTCOND */ 0)
+#include "manconf.h"
+#include "mansearch.h"
+#include "dba_array.h"
+#include "dba.h"
+
+extern const char *const mansearch_keynames[];
 
 enum   op {
        OP_DEFAULT = 0, /* new dbs from dir list or default config */
 
 enum   op {
        OP_DEFAULT = 0, /* new dbs from dir list or default config */
@@ -68,123 +67,128 @@ enum      op {
        OP_TEST /* change no databases, report potential problems */
 };
 
        OP_TEST /* change no databases, report potential problems */
 };
 
-enum   form {
-       FORM_SRC, /* format is -man or -mdoc */
-       FORM_CAT, /* format is cat */
-       FORM_NONE /* format is unknown */
-};
-
 struct str {
 struct str {
-       char            *utf8; /* key in UTF-8 form */
-       const struct of *of; /* if set, the owning parse */
-       struct str      *next; /* next in owning parse sequence */
+       const struct mpage *mpage; /* if set, the owning parse */
        uint64_t         mask; /* bitmask in sequence */
        uint64_t         mask; /* bitmask in sequence */
-       char             key[1]; /* the string itself */
+       char             key[]; /* rendered text */
 };
 
 };
 
-struct id {
-       ino_t            ino;
-       dev_t            dev;
+struct inodev {
+       ino_t            st_ino;
+       dev_t            st_dev;
 };
 
 };
 
-struct of {
-       struct id        id; /* used for hashing routine */
-       struct of       *next; /* next in ofs */
-       enum form        dform; /* path-cued form */
-       enum form        sform; /* suffix-cued form */
-       char             file[MAXPATHLEN]; /* filename rel. to manpath */
-       const char      *desc; /* parsed description */
-       const char      *sec; /* suffix-cued section (or empty) */
-       const char      *dsec; /* path-cued section (or empty) */
-       const char      *arch; /* path-cued arch. (or empty) */
-       const char      *name; /* name (from filename) (not empty) */
+struct mpage {
+       struct inodev    inodev;  /* used for hashing routine */
+       struct dba_array *dba;
+       char            *sec;     /* section from file content */
+       char            *arch;    /* architecture from file content */
+       char            *title;   /* title from file content */
+       char            *desc;    /* description from file content */
+       struct mpage    *next;    /* singly linked list */
+       struct mlink    *mlinks;  /* singly linked list */
+       int              name_head_done;
+       enum form        form;    /* format from file content */
 };
 
 };
 
-enum   stmt {
-       STMT_DELETE = 0, /* delete manpage */
-       STMT_INSERT_DOC, /* insert manpage */
-       STMT_INSERT_KEY, /* insert parsed key */
-       STMT__MAX
+struct mlink {
+       char             file[PATH_MAX]; /* filename rel. to manpath */
+       char            *dsec;    /* section from directory */
+       char            *arch;    /* architecture from directory */
+       char            *name;    /* name from file name (not empty) */
+       char            *fsec;    /* section from file name suffix */
+       struct mlink    *next;    /* singly linked list */
+       struct mpage    *mpage;   /* parent */
+       int              gzip;    /* filename has a .gz suffix */
+       enum form        dform;   /* format from directory */
+       enum form        fform;   /* format from file name suffix */
 };
 
 };
 
-typedef        int (*mdoc_fp)(struct of *, const struct mdoc_node *);
+typedef        int (*mdoc_fp)(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
 
 struct mdoc_handler {
        mdoc_fp          fp; /* optional handler */
        uint64_t         mask;  /* set unless handler returns 0 */
 
 struct mdoc_handler {
        mdoc_fp          fp; /* optional handler */
        uint64_t         mask;  /* set unless handler returns 0 */
-       int              flags;  /* for use by pmdoc_node */
-#define        MDOCF_CHILD      0x01  /* automatically index child nodes */
+       int              taboo;  /* node flags that must not be set */
 };
 
 };
 
-static void     dbclose(const char *, int);
-static void     dbindex(struct mchars *, int,
-                       const struct of *, const char *);
-static int      dbopen(const char *, int);
-static void     dbprune(const char *);
-static void     fileadd(struct of *);
-static int      filecheck(const char *);
-static void     filescan(const char *, const char *);
-static struct str *hashget(const char *, size_t);
-static void    *hash_alloc(size_t, void *);
-static void     hash_free(void *, size_t, void *);
-static void    *hash_halloc(size_t, void *);
-static void     inoadd(const struct stat *, struct of *);
-static int      inocheck(const struct stat *);
-static void     ofadd(const char *, int, const char *, 
-                       const char *, const char *, const char *, 
-                       const char *, const struct stat *);
-static void     offree(void);
-static int      ofmerge(struct mchars *, struct mparse *, const char *);
-static void     parse_catpage(struct of *, const char *);
-static int      parse_man(struct of *, 
-                       const struct man_node *);
-static void     parse_mdoc(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_body(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_head(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_Fd(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_Fn(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_In(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_Nd(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_Nm(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_Sh(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_St(struct of *, const struct mdoc_node *);
-static int      parse_mdoc_Xr(struct of *, const struct mdoc_node *);
-static int      path_reset(const char *, int, const char *);
-static void     putkey(const struct of *, 
-                       const char *, uint64_t);
-static void     putkeys(const struct of *, 
-                       const char *, int, uint64_t);
-static void     putmdockey(const struct of *,
-                       const struct mdoc_node *, uint64_t);
-static char    *stradd(const char *);
-static char    *straddbuf(const char *, size_t);
-static int      treescan(const char *);
+
+int             mandocdb(int, char *[]);
+
+static void     dbadd(struct dba *, struct mpage *);
+static void     dbadd_mlink(const struct mlink *mlink);
+static void     dbprune(struct dba *);
+static void     dbwrite(struct dba *);
+static void     filescan(const char *);
+#if HAVE_FTS_COMPARE_CONST
+static int      fts_compare(const FTSENT *const *, const FTSENT *const *);
+#else
+static int      fts_compare(const FTSENT **, const FTSENT **);
+#endif
+static void     mlink_add(struct mlink *, const struct stat *);
+static void     mlink_check(struct mpage *, struct mlink *);
+static void     mlink_free(struct mlink *);
+static void     mlinks_undupe(struct mpage *);
+static void     mpages_free(void);
+static void     mpages_merge(struct dba *, struct mparse *);
+static void     parse_cat(struct mpage *, int);
+static void     parse_man(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static void     parse_mdoc(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_head(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_Fd(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static void     parse_mdoc_fname(struct mpage *, const struct roff_node *);
+static int      parse_mdoc_Fn(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_Fo(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_Nd(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_Nm(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_Sh(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_Va(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static int      parse_mdoc_Xr(struct mpage *, const struct roff_meta *,
+                       const struct roff_node *);
+static void     putkey(const struct mpage *, char *, uint64_t);
+static void     putkeys(const struct mpage *, char *, size_t, uint64_t);
+static void     putmdockey(const struct mpage *,
+                       const struct roff_node *, uint64_t, int);
+static int      render_string(char **, size_t *);
+static void     say(const char *, const char *, ...)
+                       __attribute__((__format__ (__printf__, 2, 3)));
+static int      set_basedir(const char *, int);
+static int      treescan(void);
 static size_t   utf8(unsigned int, char [7]);
 static size_t   utf8(unsigned int, char [7]);
-static void     utf8key(struct mchars *, struct str *);
-static void     wordaddbuf(const struct of *, 
-                       const char *, size_t, uint64_t);
 
 
-static char            *progname;
-static int              use_all; /* use all found files */
 static int              nodb; /* no database changes */
 static int              nodb; /* no database changes */
-static int              verb; /* print what we're doing */
-static int              warnings; /* warn about crap */
-static enum op          op; /* operational mode */
-static struct ohash     inos; /* table of inodes/devices */
-static struct ohash     filenames; /* table of filenames */
+static int              mparse_options; /* abort the parse early */
+static int              use_all; /* use all found files */
+static int              debug; /* print what we're doing */
+static int              warnings; /* warn about crap */
+static int              write_utf8; /* write UTF-8 output; else ASCII */
+static int              exitcode; /* to be returned by main */
+static enum op          op; /* operational mode */
+static char             basedir[PATH_MAX]; /* current base directory */
+static struct mpage    *mpage_head; /* list of distinct manual pages */
+static struct ohash     mpages; /* table of distinct manual pages */
+static struct ohash     mlinks; /* table of directory entries */
+static struct ohash     names; /* table of all names */
 static struct ohash     strings; /* table of all strings */
 static struct ohash     strings; /* table of all strings */
-static struct of       *ofs = NULL; /* vector of files to parse */
-static struct str      *words = NULL; /* word list in current parse */
-static sqlite3         *db = NULL; /* current database */
-static sqlite3_stmt    *stmts[STMT__MAX]; /* current statements */
-
-static const struct mdoc_handler mdocs[MDOC_MAX] = {
-       { NULL, 0, 0 },  /* Ap */
-       { NULL, 0, 0 },  /* Dd */
-       { NULL, 0, 0 },  /* Dt */
-       { NULL, 0, 0 },  /* Os */
-       { parse_mdoc_Sh, TYPE_Sh, MDOCF_CHILD }, /* Sh */
-       { parse_mdoc_head, TYPE_Ss, MDOCF_CHILD }, /* Ss */
+static uint64_t         name_mask;
+
+static const struct mdoc_handler __mdocs[MDOC_MAX - MDOC_Dd] = {
+       { NULL, 0, NODE_NOPRT },  /* Dd */
+       { NULL, 0, NODE_NOPRT },  /* Dt */
+       { NULL, 0, NODE_NOPRT },  /* Os */
+       { parse_mdoc_Sh, TYPE_Sh, 0 }, /* Sh */
+       { parse_mdoc_head, TYPE_Ss, 0 }, /* Ss */
        { NULL, 0, 0 },  /* Pp */
        { NULL, 0, 0 },  /* D1 */
        { NULL, 0, 0 },  /* Dl */
        { NULL, 0, 0 },  /* Pp */
        { NULL, 0, 0 },  /* D1 */
        { NULL, 0, 0 },  /* Dl */
@@ -194,32 +198,33 @@ static    const struct mdoc_handler mdocs[MDOC_MAX] = {
        { NULL, 0, 0 },  /* El */
        { NULL, 0, 0 },  /* It */
        { NULL, 0, 0 },  /* Ad */
        { NULL, 0, 0 },  /* El */
        { NULL, 0, 0 },  /* It */
        { NULL, 0, 0 },  /* Ad */
-       { NULL, TYPE_An, MDOCF_CHILD },  /* An */
-       { NULL, TYPE_Ar, MDOCF_CHILD },  /* Ar */
-       { NULL, TYPE_Cd, MDOCF_CHILD },  /* Cd */
-       { NULL, TYPE_Cm, MDOCF_CHILD },  /* Cm */
-       { NULL, TYPE_Dv, MDOCF_CHILD },  /* Dv */
-       { NULL, TYPE_Er, MDOCF_CHILD },  /* Er */
-       { NULL, TYPE_Ev, MDOCF_CHILD },  /* Ev */
+       { NULL, TYPE_An, 0 },  /* An */
+       { NULL, 0, 0 },  /* Ap */
+       { NULL, TYPE_Ar, 0 },  /* Ar */
+       { NULL, TYPE_Cd, 0 },  /* Cd */
+       { NULL, TYPE_Cm, 0 },  /* Cm */
+       { NULL, TYPE_Dv, 0 },  /* Dv */
+       { NULL, TYPE_Er, 0 },  /* Er */
+       { NULL, TYPE_Ev, 0 },  /* Ev */
        { NULL, 0, 0 },  /* Ex */
        { NULL, 0, 0 },  /* Ex */
-       { NULL, TYPE_Fa, MDOCF_CHILD },  /* Fa */
-       { parse_mdoc_Fd, TYPE_In, 0 },  /* Fd */
-       { NULL, TYPE_Fl, MDOCF_CHILD },  /* Fl */
+       { NULL, TYPE_Fa, 0 },  /* Fa */
+       { parse_mdoc_Fd, 0, 0 },  /* Fd */
+       { NULL, TYPE_Fl, 0 },  /* Fl */
        { parse_mdoc_Fn, 0, 0 },  /* Fn */
        { parse_mdoc_Fn, 0, 0 },  /* Fn */
-       { NULL, TYPE_Ft, MDOCF_CHILD },  /* Ft */
-       { NULL, TYPE_Ic, MDOCF_CHILD },  /* Ic */
-       { parse_mdoc_In, TYPE_In, MDOCF_CHILD },  /* In */
-       { NULL, TYPE_Li, MDOCF_CHILD },  /* Li */
-       { parse_mdoc_Nd, TYPE_Nd, MDOCF_CHILD },  /* Nd */
-       { parse_mdoc_Nm, TYPE_Nm, MDOCF_CHILD },  /* Nm */
+       { NULL, TYPE_Ft, 0 },  /* Ft */
+       { NULL, TYPE_Ic, 0 },  /* Ic */
+       { NULL, TYPE_In, 0 },  /* In */
+       { NULL, TYPE_Li, 0 },  /* Li */
+       { parse_mdoc_Nd, 0, 0 },  /* Nd */
+       { parse_mdoc_Nm, 0, 0 },  /* Nm */
        { NULL, 0, 0 },  /* Op */
        { NULL, 0, 0 },  /* Ot */
        { NULL, 0, 0 },  /* Op */
        { NULL, 0, 0 },  /* Ot */
-       { NULL, TYPE_Pa, MDOCF_CHILD },  /* Pa */
+       { NULL, TYPE_Pa, NODE_NOSRC },  /* Pa */
        { NULL, 0, 0 },  /* Rv */
        { NULL, 0, 0 },  /* Rv */
-       { parse_mdoc_St, TYPE_St, 0 },  /* St */
-       { NULL, TYPE_Va, MDOCF_CHILD },  /* Va */
-       { parse_mdoc_body, TYPE_Va, MDOCF_CHILD },  /* Vt */
-       { parse_mdoc_Xr, TYPE_Xr, 0 },  /* Xr */
+       { NULL, TYPE_St, 0 },  /* St */
+       { parse_mdoc_Va, TYPE_Va, 0 },  /* Va */
+       { parse_mdoc_Va, TYPE_Vt, 0 },  /* Vt */
+       { parse_mdoc_Xr, 0, 0 },  /* Xr */
        { NULL, 0, 0 },  /* %A */
        { NULL, 0, 0 },  /* %B */
        { NULL, 0, 0 },  /* %D */
        { NULL, 0, 0 },  /* %A */
        { NULL, 0, 0 },  /* %B */
        { NULL, 0, 0 },  /* %D */
@@ -234,27 +239,27 @@ static    const struct mdoc_handler mdocs[MDOC_MAX] = {
        { NULL, 0, 0 },  /* Ac */
        { NULL, 0, 0 },  /* Ao */
        { NULL, 0, 0 },  /* Aq */
        { NULL, 0, 0 },  /* Ac */
        { NULL, 0, 0 },  /* Ao */
        { NULL, 0, 0 },  /* Aq */
-       { NULL, TYPE_At, MDOCF_CHILD },  /* At */
+       { NULL, TYPE_At, 0 },  /* At */
        { NULL, 0, 0 },  /* Bc */
        { NULL, 0, 0 },  /* Bf */
        { NULL, 0, 0 },  /* Bo */
        { NULL, 0, 0 },  /* Bq */
        { NULL, 0, 0 },  /* Bc */
        { NULL, 0, 0 },  /* Bf */
        { NULL, 0, 0 },  /* Bo */
        { NULL, 0, 0 },  /* Bq */
-       { NULL, TYPE_Bsx, MDOCF_CHILD },  /* Bsx */
-       { NULL, TYPE_Bx, MDOCF_CHILD },  /* Bx */
+       { NULL, TYPE_Bsx, NODE_NOSRC },  /* Bsx */
+       { NULL, TYPE_Bx, NODE_NOSRC },  /* Bx */
        { NULL, 0, 0 },  /* Db */
        { NULL, 0, 0 },  /* Dc */
        { NULL, 0, 0 },  /* Do */
        { NULL, 0, 0 },  /* Dq */
        { NULL, 0, 0 },  /* Ec */
        { NULL, 0, 0 },  /* Ef */
        { NULL, 0, 0 },  /* Db */
        { NULL, 0, 0 },  /* Dc */
        { NULL, 0, 0 },  /* Do */
        { NULL, 0, 0 },  /* Dq */
        { NULL, 0, 0 },  /* Ec */
        { NULL, 0, 0 },  /* Ef */
-       { NULL, TYPE_Em, MDOCF_CHILD },  /* Em */
+       { NULL, TYPE_Em, 0 },  /* Em */
        { NULL, 0, 0 },  /* Eo */
        { NULL, 0, 0 },  /* Eo */
-       { NULL, TYPE_Fx, MDOCF_CHILD },  /* Fx */
-       { NULL, TYPE_Ms, MDOCF_CHILD },  /* Ms */
+       { NULL, TYPE_Fx, NODE_NOSRC },  /* Fx */
+       { NULL, TYPE_Ms, 0 },  /* Ms */
        { NULL, 0, 0 },  /* No */
        { NULL, 0, 0 },  /* Ns */
        { NULL, 0, 0 },  /* No */
        { NULL, 0, 0 },  /* Ns */
-       { NULL, TYPE_Nx, MDOCF_CHILD },  /* Nx */
-       { NULL, TYPE_Ox, MDOCF_CHILD },  /* Ox */
+       { NULL, TYPE_Nx, NODE_NOSRC },  /* Nx */
+       { NULL, TYPE_Ox, NODE_NOSRC },  /* Ox */
        { NULL, 0, 0 },  /* Pc */
        { NULL, 0, 0 },  /* Pf */
        { NULL, 0, 0 },  /* Po */
        { NULL, 0, 0 },  /* Pc */
        { NULL, 0, 0 },  /* Pf */
        { NULL, 0, 0 },  /* Po */
@@ -270,12 +275,12 @@ static    const struct mdoc_handler mdocs[MDOC_MAX] = {
        { NULL, 0, 0 },  /* Sq */
        { NULL, 0, 0 },  /* Sm */
        { NULL, 0, 0 },  /* Sx */
        { NULL, 0, 0 },  /* Sq */
        { NULL, 0, 0 },  /* Sm */
        { NULL, 0, 0 },  /* Sx */
-       { NULL, TYPE_Sy, MDOCF_CHILD },  /* Sy */
-       { NULL, TYPE_Tn, MDOCF_CHILD },  /* Tn */
-       { NULL, 0, 0 },  /* Ux */
+       { NULL, TYPE_Sy, 0 },  /* Sy */
+       { NULL, TYPE_Tn, 0 },  /* Tn */
+       { NULL, 0, NODE_NOSRC },  /* Ux */
        { NULL, 0, 0 },  /* Xc */
        { NULL, 0, 0 },  /* Xo */
        { NULL, 0, 0 },  /* Xc */
        { NULL, 0, 0 },  /* Xo */
-       { parse_mdoc_head, TYPE_Fn, 0 },  /* Fo */
+       { parse_mdoc_Fo, 0, 0 },  /* Fo */
        { NULL, 0, 0 },  /* Fc */
        { NULL, 0, 0 },  /* Oo */
        { NULL, 0, 0 },  /* Oc */
        { NULL, 0, 0 },  /* Fc */
        { NULL, 0, 0 },  /* Oo */
        { NULL, 0, 0 },  /* Oc */
@@ -285,119 +290,112 @@ static  const struct mdoc_handler mdocs[MDOC_MAX] = {
        { NULL, 0, 0 },  /* Hf */
        { NULL, 0, 0 },  /* Fr */
        { NULL, 0, 0 },  /* Ud */
        { NULL, 0, 0 },  /* Hf */
        { NULL, 0, 0 },  /* Fr */
        { NULL, 0, 0 },  /* Ud */
-       { NULL, TYPE_Lb, MDOCF_CHILD },  /* Lb */
+       { NULL, TYPE_Lb, NODE_NOSRC },  /* Lb */
        { NULL, 0, 0 },  /* Lp */
        { NULL, 0, 0 },  /* Lp */
-       { NULL, TYPE_Lk, MDOCF_CHILD },  /* Lk */
-       { NULL, TYPE_Mt, MDOCF_CHILD },  /* Mt */
+       { NULL, TYPE_Lk, 0 },  /* Lk */
+       { NULL, TYPE_Mt, NODE_NOSRC },  /* Mt */
        { NULL, 0, 0 },  /* Brq */
        { NULL, 0, 0 },  /* Bro */
        { NULL, 0, 0 },  /* Brc */
        { NULL, 0, 0 },  /* %C */
        { NULL, 0, 0 },  /* Es */
        { NULL, 0, 0 },  /* En */
        { NULL, 0, 0 },  /* Brq */
        { NULL, 0, 0 },  /* Bro */
        { NULL, 0, 0 },  /* Brc */
        { NULL, 0, 0 },  /* %C */
        { NULL, 0, 0 },  /* Es */
        { NULL, 0, 0 },  /* En */
-       { NULL, TYPE_Dx, MDOCF_CHILD },  /* Dx */
+       { NULL, TYPE_Dx, NODE_NOSRC },  /* Dx */
        { NULL, 0, 0 },  /* %Q */
        { NULL, 0, 0 },  /* %Q */
-       { NULL, 0, 0 },  /* br */
-       { NULL, 0, 0 },  /* sp */
        { NULL, 0, 0 },  /* %U */
        { NULL, 0, 0 },  /* Ta */
 };
        { NULL, 0, 0 },  /* %U */
        { NULL, 0, 0 },  /* Ta */
 };
+static const struct mdoc_handler *const mdocs = __mdocs - MDOC_Dd;
+
 
 int
 
 int
-main(int argc, char *argv[])
+mandocdb(int argc, char *argv[])
 {
 {
-       char              cwd[MAXPATHLEN];
-       int               ch, rc, fd, i;
-       size_t            j, sz;
-       const char       *dir;
-       struct str       *s;
-       struct mchars    *mc;
-       struct manpaths   dirs;
+       struct manconf    conf;
        struct mparse    *mp;
        struct mparse    *mp;
-       struct ohash_info ino_info, filename_info, str_info;
-
-       memset(stmts, 0, STMT__MAX * sizeof(sqlite3_stmt *));
-       memset(&dirs, 0, sizeof(struct manpaths));
-
-       ino_info.halloc = filename_info.halloc = 
-               str_info.halloc = hash_halloc;
-       ino_info.hfree = filename_info.hfree = 
-               str_info.hfree = hash_free;
-       ino_info.alloc = filename_info.alloc = 
-               str_info.alloc = hash_alloc;
-
-       ino_info.key_offset = offsetof(struct of, id);
-       filename_info.key_offset = offsetof(struct of, file);
-       str_info.key_offset = offsetof(struct str, key);
+       struct dba       *dba;
+       const char       *path_arg, *progname;
+       size_t            j, sz;
+       int               ch, i;
 
 
-       progname = strrchr(argv[0], '/');
-       if (progname == NULL)
-               progname = argv[0];
-       else
-               ++progname;
+#if HAVE_PLEDGE
+       if (pledge("stdio rpath wpath cpath fattr flock proc exec", NULL) == -1) {
+               warn("pledge");
+               return (int)MANDOCLEVEL_SYSERR;
+       }
+#endif
 
 
-       /*
-        * Remember where we started by keeping a fd open to the origin
-        * path component: throughout this utility, we chdir() a lot to
-        * handle relative paths, and by doing this, we can return to
-        * the starting point.
-        */
-       if (NULL == getcwd(cwd, MAXPATHLEN)) {
-               perror(NULL);
-               return(EXIT_FAILURE);
-       } else if (-1 == (fd = open(cwd, O_RDONLY, 0))) {
-               perror(cwd);
-               return(EXIT_FAILURE);
+#if HAVE_SANDBOX_INIT
+       if (sandbox_init(kSBXProfileNoInternet, SANDBOX_NAMED, NULL) == -1) {
+               warnx("sandbox_init");
+               return (int)MANDOCLEVEL_SYSERR;
        }
        }
+#endif
+
+       memset(&conf, 0, sizeof(conf));
 
        /*
 
        /*
-        * We accept a few different invocations.  
+        * We accept a few different invocations.
         * The CHECKOP macro makes sure that invocation styles don't
         * clobber each other.
         */
 #define        CHECKOP(_op, _ch) do \
        if (OP_DEFAULT != (_op)) { \
         * The CHECKOP macro makes sure that invocation styles don't
         * clobber each other.
         */
 #define        CHECKOP(_op, _ch) do \
        if (OP_DEFAULT != (_op)) { \
-               fprintf(stderr, "-%c: Conflicting option\n", (_ch)); \
+               warnx("-%c: Conflicting option", (_ch)); \
                goto usage; \
        } while (/*CONSTCOND*/0)
 
                goto usage; \
        } while (/*CONSTCOND*/0)
 
-       dir = NULL;
+       path_arg = NULL;
        op = OP_DEFAULT;
 
        op = OP_DEFAULT;
 
-       while (-1 != (ch = getopt(argc, argv, "aC:d:ntu:vW")))
+       while (-1 != (ch = getopt(argc, argv, "aC:Dd:npQT:tu:v")))
                switch (ch) {
                switch (ch) {
-               case ('a'):
+               case 'a':
                        use_all = 1;
                        break;
                        use_all = 1;
                        break;
-               case ('C'):
+               case 'C':
                        CHECKOP(op, ch);
                        CHECKOP(op, ch);
-                       dir = optarg;
+                       path_arg = optarg;
                        op = OP_CONFFILE;
                        break;
                        op = OP_CONFFILE;
                        break;
-               case ('d'):
+               case 'D':
+                       debug++;
+                       break;
+               case 'd':
                        CHECKOP(op, ch);
                        CHECKOP(op, ch);
-                       dir = optarg;
+                       path_arg = optarg;
                        op = OP_UPDATE;
                        break;
                        op = OP_UPDATE;
                        break;
-               case ('n'):
+               case 'n':
                        nodb = 1;
                        break;
                        nodb = 1;
                        break;
-               case ('t'):
+               case 'p':
+                       warnings = 1;
+                       break;
+               case 'Q':
+                       mparse_options |= MPARSE_QUICK;
+                       break;
+               case 'T':
+                       if (strcmp(optarg, "utf8")) {
+                               warnx("-T%s: Unsupported output format",
+                                   optarg);
+                               goto usage;
+                       }
+                       write_utf8 = 1;
+                       break;
+               case 't':
                        CHECKOP(op, ch);
                        dup2(STDOUT_FILENO, STDERR_FILENO);
                        op = OP_TEST;
                        nodb = warnings = 1;
                        break;
                        CHECKOP(op, ch);
                        dup2(STDOUT_FILENO, STDERR_FILENO);
                        op = OP_TEST;
                        nodb = warnings = 1;
                        break;
-               case ('u'):
+               case 'u':
                        CHECKOP(op, ch);
                        CHECKOP(op, ch);
-                       dir = optarg;
+                       path_arg = optarg;
                        op = OP_DELETE;
                        break;
                        op = OP_DELETE;
                        break;
-               case ('v'):
-                       verb++;
-                       break;
-               case ('W'):
-                       warnings = 1;
+               case 'v':
+                       /* Compatibility with espie@'s makewhatis. */
                        break;
                default:
                        goto usage;
                        break;
                default:
                        goto usage;
@@ -406,301 +404,485 @@ main(int argc, char *argv[])
        argc -= optind;
        argv += optind;
 
        argc -= optind;
        argv += optind;
 
+#if HAVE_PLEDGE
+       if (nodb) {
+               if (pledge("stdio rpath", NULL) == -1) {
+                       warn("pledge");
+                       return (int)MANDOCLEVEL_SYSERR;
+               }
+       }
+#endif
+
        if (OP_CONFFILE == op && argc > 0) {
        if (OP_CONFFILE == op && argc > 0) {
-               fprintf(stderr, "-C: Too many arguments\n");
+               warnx("-C: Too many arguments");
                goto usage;
        }
 
                goto usage;
        }
 
-       rc = 1;
-       mp = mparse_alloc(MPARSE_AUTO, 
-               MANDOCLEVEL_FATAL, NULL, NULL, NULL);
-       mc = mchars_alloc();
-
-       ohash_init(&strings, 6, &str_info);
-       ohash_init(&inos, 6, &ino_info);
-       ohash_init(&filenames, 6, &filename_info);
+       exitcode = (int)MANDOCLEVEL_OK;
+       mchars_alloc();
+       mp = mparse_alloc(mparse_options, MANDOCLEVEL_BADARG, NULL,
+           MANDOC_OS_OTHER, NULL);
+       mandoc_ohash_init(&mpages, 6, offsetof(struct mpage, inodev));
+       mandoc_ohash_init(&mlinks, 6, offsetof(struct mlink, file));
 
        if (OP_UPDATE == op || OP_DELETE == op || OP_TEST == op) {
 
        if (OP_UPDATE == op || OP_DELETE == op || OP_TEST == op) {
-               /* 
-                * Force processing all files.
-                */
-               use_all = 1;
-               if (NULL == dir)
-                       dir = cwd;
+
                /*
                /*
-                * All of these deal with a specific directory.
-                * Jump into that directory then collect files specified
-                * on the command-line.
+                * Most of these deal with a specific directory.
+                * Jump into that directory first.
                 */
                 */
-               if (0 == path_reset(cwd, fd, dir))
-                       goto out;
-               for (i = 0; i < argc; i++)
-                       filescan(argv[i], dir);
-               if (0 == dbopen(dir, 1))
+               if (OP_TEST != op && 0 == set_basedir(path_arg, 1))
                        goto out;
                        goto out;
-               if (OP_TEST != op)
-                       dbprune(dir);
+
+               dba = nodb ? dba_new(128) : dba_read(MANDOC_DB);
+               if (dba != NULL) {
+                       /*
+                        * The existing database is usable.  Process
+                        * all files specified on the command-line.
+                        */
+#if HAVE_PLEDGE
+                       if (!nodb) {
+                               if (pledge("stdio rpath wpath cpath fattr flock", NULL) == -1) {
+                                       warn("pledge");
+                                       exitcode = (int)MANDOCLEVEL_SYSERR;
+                                       goto out;
+                               }
+                       }
+#endif
+                       use_all = 1;
+                       for (i = 0; i < argc; i++)
+                               filescan(argv[i]);
+                       if (nodb == 0)
+                               dbprune(dba);
+               } else {
+                       /* Database missing or corrupt. */
+                       if (op != OP_UPDATE || errno != ENOENT)
+                               say(MANDOC_DB, "%s: Automatically recreating"
+                                   " from scratch", strerror(errno));
+                       exitcode = (int)MANDOCLEVEL_OK;
+                       op = OP_DEFAULT;
+                       if (0 == treescan())
+                               goto out;
+                       dba = dba_new(128);
+               }
                if (OP_DELETE != op)
                if (OP_DELETE != op)
-                       rc = ofmerge(mc, mp, dir);
-               dbclose(dir, 1);
+                       mpages_merge(dba, mp);
+               if (nodb == 0)
+                       dbwrite(dba);
+               dba_free(dba);
        } else {
                /*
                 * If we have arguments, use them as our manpaths.
        } else {
                /*
                 * If we have arguments, use them as our manpaths.
-                * If we don't, grok from manpath(1) or however else
-                * manpath_parse() wants to do it.
+                * If we don't, use man.conf(5).
                 */
                if (argc > 0) {
                 */
                if (argc > 0) {
-                       dirs.paths = mandoc_calloc
-                               (argc, sizeof(char *));
-                       dirs.sz = (size_t)argc;
+                       conf.manpath.paths = mandoc_reallocarray(NULL,
+                           argc, sizeof(char *));
+                       conf.manpath.sz = (size_t)argc;
                        for (i = 0; i < argc; i++)
                        for (i = 0; i < argc; i++)
-                               dirs.paths[i] = mandoc_strdup(argv[i]);
+                               conf.manpath.paths[i] = mandoc_strdup(argv[i]);
                } else
                } else
-                       manpath_parse(&dirs, dir, NULL, NULL);
+                       manconf_parse(&conf, path_arg, NULL, NULL);
+
+               if (conf.manpath.sz == 0) {
+                       exitcode = (int)MANDOCLEVEL_BADARG;
+                       say("", "Empty manpath");
+               }
 
                /*
 
                /*
-                * First scan the tree rooted at a base directory.
-                * Then whak its database (if one exists), parse, and
-                * build up the database.
+                * First scan the tree rooted at a base directory, then
+                * build a new database and finally move it into place.
                 * Ignore zero-length directories and strip trailing
                 * slashes.
                 */
                 * Ignore zero-length directories and strip trailing
                 * slashes.
                 */
-               for (j = 0; j < dirs.sz; j++) {
-                       sz = strlen(dirs.paths[j]);
-                       if (sz && '/' == dirs.paths[j][sz - 1])
-                               dirs.paths[j][--sz] = '\0';
+               for (j = 0; j < conf.manpath.sz; j++) {
+                       sz = strlen(conf.manpath.paths[j]);
+                       if (sz && conf.manpath.paths[j][sz - 1] == '/')
+                               conf.manpath.paths[j][--sz] = '\0';
                        if (0 == sz)
                                continue;
                        if (0 == sz)
                                continue;
-                       if (0 == path_reset(cwd, fd, dirs.paths[j]))
-                               goto out;
-                       if (0 == treescan(dirs.paths[j]))
-                               goto out;
-                       if (0 == path_reset(cwd, fd, dirs.paths[j]))
-                               goto out;
-                       if (0 == dbopen(dirs.paths[j], 0))
-                               goto out;
-                       if (0 == ofmerge(mc, mp, dirs.paths[j]))
-                               goto out;
-                       dbclose(dirs.paths[j], 0);
-                       offree();
-                       ohash_delete(&inos);
-                       ohash_init(&inos, 6, &ino_info);
-                       ohash_delete(&filenames);
-                       ohash_init(&filenames, 6, &filename_info);
+
+                       if (j) {
+                               mandoc_ohash_init(&mpages, 6,
+                                   offsetof(struct mpage, inodev));
+                               mandoc_ohash_init(&mlinks, 6,
+                                   offsetof(struct mlink, file));
+                       }
+
+                       if ( ! set_basedir(conf.manpath.paths[j], argc > 0))
+                               continue;
+                       if (0 == treescan())
+                               continue;
+                       dba = dba_new(128);
+                       mpages_merge(dba, mp);
+                       if (nodb == 0)
+                               dbwrite(dba);
+                       dba_free(dba);
+
+                       if (j + 1 < conf.manpath.sz) {
+                               mpages_free();
+                               ohash_delete(&mpages);
+                               ohash_delete(&mlinks);
+                       }
                }
        }
 out:
                }
        }
 out:
-       close(fd);
-       manpath_free(&dirs);
-       mchars_free(mc);
+       manconf_free(&conf);
        mparse_free(mp);
        mparse_free(mp);
-       for (s = ohash_first(&strings, &ch);
-                       NULL != s; s = ohash_next(&strings, &ch)) {
-               if (s->utf8 != s->key)
-                       free(s->utf8);
-               free(s);
-       }
-       ohash_delete(&strings);
-       ohash_delete(&inos);
-       ohash_delete(&filenames);
-       offree();
-       return(rc ? EXIT_SUCCESS : EXIT_FAILURE);
+       mchars_free();
+       mpages_free();
+       ohash_delete(&mpages);
+       ohash_delete(&mlinks);
+       return exitcode;
 usage:
 usage:
-       fprintf(stderr, "usage: %s [-anvW] [-C file]\n"
-                       "       %s [-anvW] dir ...\n"
-                       "       %s [-nvW] -d dir [file ...]\n"
-                       "       %s [-nvW] -u dir [file ...]\n"
-                       "       %s -t file ...\n",
-                      progname, progname, progname, 
-                      progname, progname);
-
-       return(EXIT_FAILURE);
+       progname = getprogname();
+       fprintf(stderr, "usage: %s [-aDnpQ] [-C file] [-Tutf8]\n"
+                       "       %s [-aDnpQ] [-Tutf8] dir ...\n"
+                       "       %s [-DnpQ] [-Tutf8] -d dir [file ...]\n"
+                       "       %s [-Dnp] -u dir [file ...]\n"
+                       "       %s [-Q] -t file ...\n",
+                       progname, progname, progname, progname, progname);
+
+       return (int)MANDOCLEVEL_BADARG;
 }
 
 /*
 }
 
 /*
- * Scan a directory tree rooted at "base" for manpages.
+ * To get a singly linked list in alpha order while inserting entries
+ * at the beginning, process directory entries in reverse alpha order.
+ */
+static int
+#if HAVE_FTS_COMPARE_CONST
+fts_compare(const FTSENT *const *a, const FTSENT *const *b)
+#else
+fts_compare(const FTSENT **a, const FTSENT **b)
+#endif
+{
+       return -strcmp((*a)->fts_name, (*b)->fts_name);
+}
+
+/*
+ * Scan a directory tree rooted at "basedir" for manpages.
  * We use fts(), scanning directory parts along the way for clues to our
  * section and architecture.
  *
  * If use_all has been specified, grok all files.
  * If not, sanitise paths to the following:
  *
  * We use fts(), scanning directory parts along the way for clues to our
  * section and architecture.
  *
  * If use_all has been specified, grok all files.
  * If not, sanitise paths to the following:
  *
- *   [./]man*[/<arch>]/<name>.<section> 
+ *   [./]man*[/<arch>]/<name>.<section>
  *   or
  *   [./]cat<section>[/<arch>]/<name>.0
  *
  *   or
  *   [./]cat<section>[/<arch>]/<name>.0
  *
- * TODO: accomodate for multi-language directories.
+ * TODO: accommodate for multi-language directories.
  */
 static int
  */
 static int
-treescan(const char *base)
+treescan(void)
 {
 {
+       char             buf[PATH_MAX];
        FTS             *f;
        FTSENT          *ff;
        FTS             *f;
        FTSENT          *ff;
-       int              dform;
-       char            *sec;
-       const char      *dsec, *arch, *cp, *name, *path;
+       struct mlink    *mlink;
+       int              gzip;
+       enum form        dform;
+       char            *dsec, *arch, *fsec, *cp;
+       const char      *path;
        const char      *argv[2];
 
        argv[0] = ".";
        const char      *argv[2];
 
        argv[0] = ".";
-       argv[1] = (char *)NULL;
-
-       /*
-        * Walk through all components under the directory, using the
-        * logical descent of files.
-        */
-       f = fts_open((char * const *)argv, FTS_LOGICAL, NULL);
-       if (NULL == f) {
-               perror(base);
-               return(0);
+       argv[1] = NULL;
+
+       f = fts_open((char * const *)argv, FTS_PHYSICAL | FTS_NOCHDIR,
+           fts_compare);
+       if (f == NULL) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "&fts_open");
+               return 0;
        }
 
        dsec = arch = NULL;
        dform = FORM_NONE;
 
        }
 
        dsec = arch = NULL;
        dform = FORM_NONE;
 
-       while (NULL != (ff = fts_read(f))) {
+       while ((ff = fts_read(f)) != NULL) {
                path = ff->fts_path + 2;
                path = ff->fts_path + 2;
+               switch (ff->fts_info) {
+
                /*
                /*
-                * If we're a regular file, add an "of" by using the
-                * stored directory data and handling the filename.
-                * Disallow duplicate (hard-linked) files.
+                * Symbolic links require various sanity checks,
+                * then get handled just like regular files.
                 */
                 */
-               if (FTS_F == ff->fts_info) {
-                       if ( ! use_all && ff->fts_level < 2) {
-                               WARNING(path, base, "Extraneous file");
+               case FTS_SL:
+                       if (realpath(path, buf) == NULL) {
+                               if (warnings)
+                                       say(path, "&realpath");
+                               continue;
+                       }
+                       if (strstr(buf, basedir) != buf
+#ifdef HOMEBREWDIR
+                           && strstr(buf, HOMEBREWDIR) != buf
+#endif
+                       ) {
+                               if (warnings) say("",
+                                   "%s: outside base directory", buf);
                                continue;
                                continue;
-                       } else if (inocheck(ff->fts_statp)) {
-                               WARNING(path, base, "Duplicate file");
+                       }
+                       /* Use logical inode to avoid mpages dupe. */
+                       if (stat(path, ff->fts_statp) == -1) {
+                               if (warnings)
+                                       say(path, "&stat");
                                continue;
                                continue;
-                       } 
-
-                       cp = ff->fts_name;
+                       }
+                       /* FALLTHROUGH */
 
 
-                       if (0 == strcmp(cp, "mandocdb.db")) {
-                               WARNING(path, base, "Skip database");
+               /*
+                * If we're a regular file, add an mlink by using the
+                * stored directory data and handling the filename.
+                */
+               case FTS_F:
+                       if ( ! strcmp(path, MANDOC_DB))
                                continue;
                                continue;
-                       } else if (NULL != (cp = strrchr(cp, '.'))) {
-                               if (0 == strcmp(cp + 1, "html")) {
-                                       WARNING(path, base, "Skip html");
-                                       continue;
-                               } else if (0 == strcmp(cp + 1, "gz")) {
-                                       WARNING(path, base, "Skip gz");
-                                       continue;
-                               } else if (0 == strcmp(cp + 1, "ps")) {
-                                       WARNING(path, base, "Skip ps");
-                                       continue;
-                               } else if (0 == strcmp(cp + 1, "pdf")) {
-                                       WARNING(path, base, "Skip pdf");
+                       if ( ! use_all && ff->fts_level < 2) {
+                               if (warnings)
+                                       say(path, "Extraneous file");
+                               continue;
+                       }
+                       gzip = 0;
+                       fsec = NULL;
+                       while (fsec == NULL) {
+                               fsec = strrchr(ff->fts_name, '.');
+                               if (fsec == NULL || strcmp(fsec+1, "gz"))
+                                       break;
+                               gzip = 1;
+                               *fsec = '\0';
+                               fsec = NULL;
+                       }
+                       if (fsec == NULL) {
+                               if ( ! use_all) {
+                                       if (warnings)
+                                               say(path,
+                                                   "No filename suffix");
                                        continue;
                                }
                                        continue;
                                }
+                       } else if ( ! strcmp(++fsec, "html")) {
+                               if (warnings)
+                                       say(path, "Skip html");
+                               continue;
+                       } else if ( ! strcmp(fsec, "ps")) {
+                               if (warnings)
+                                       say(path, "Skip ps");
+                               continue;
+                       } else if ( ! strcmp(fsec, "pdf")) {
+                               if (warnings)
+                                       say(path, "Skip pdf");
+                               continue;
+                       } else if ( ! use_all &&
+                           ((dform == FORM_SRC &&
+                             strncmp(fsec, dsec, strlen(dsec))) ||
+                            (dform == FORM_CAT && strcmp(fsec, "0")))) {
+                               if (warnings)
+                                       say(path, "Wrong filename suffix");
+                               continue;
+                       } else
+                               fsec[-1] = '\0';
+
+                       mlink = mandoc_calloc(1, sizeof(struct mlink));
+                       if (strlcpy(mlink->file, path,
+                           sizeof(mlink->file)) >=
+                           sizeof(mlink->file)) {
+                               say(path, "Filename too long");
+                               free(mlink);
+                               continue;
                        }
                        }
-
-                       if (NULL != (sec = strrchr(ff->fts_name, '.'))) {
-                               *sec = '\0';
-                               sec = stradd(sec + 1);
-                       }
-                       name = stradd(ff->fts_name);
-                       ofadd(base, dform, path, 
-                               name, dsec, sec, arch, ff->fts_statp);
+                       mlink->dform = dform;
+                       mlink->dsec = dsec;
+                       mlink->arch = arch;
+                       mlink->name = ff->fts_name;
+                       mlink->fsec = fsec;
+                       mlink->gzip = gzip;
+                       mlink_add(mlink, ff->fts_statp);
                        continue;
                        continue;
-               } else if (FTS_D != ff->fts_info && 
-                               FTS_DP != ff->fts_info)
+
+               case FTS_D:
+               case FTS_DP:
+                       break;
+
+               default:
+                       if (warnings)
+                               say(path, "Not a regular file");
                        continue;
                        continue;
+               }
 
                switch (ff->fts_level) {
 
                switch (ff->fts_level) {
-               case (0):
+               case 0:
                        /* Ignore the root directory. */
                        break;
                        /* Ignore the root directory. */
                        break;
-               case (1):
+               case 1:
                        /*
                         * This might contain manX/ or catX/.
                         * Try to infer this from the name.
                         * If we're not in use_all, enforce it.
                         */
                        /*
                         * This might contain manX/ or catX/.
                         * Try to infer this from the name.
                         * If we're not in use_all, enforce it.
                         */
-                       dsec = NULL;
-                       dform = FORM_NONE;
                        cp = ff->fts_name;
                        cp = ff->fts_name;
-                       if (FTS_DP == ff->fts_info)
+                       if (ff->fts_info == FTS_DP) {
+                               dform = FORM_NONE;
+                               dsec = NULL;
                                break;
                                break;
+                       }
 
 
-                       if (0 == strncmp(cp, "man", 3)) {
+                       if ( ! strncmp(cp, "man", 3)) {
                                dform = FORM_SRC;
                                dform = FORM_SRC;
-                               dsec = stradd(cp + 3);
-                       } else if (0 == strncmp(cp, "cat", 3)) {
+                               dsec = cp + 3;
+                       } else if ( ! strncmp(cp, "cat", 3)) {
                                dform = FORM_CAT;
                                dform = FORM_CAT;
-                               dsec = stradd(cp + 3);
+                               dsec = cp + 3;
+                       } else {
+                               dform = FORM_NONE;
+                               dsec = NULL;
                        }
 
                        }
 
-                       if (NULL != dsec || use_all) 
+                       if (dsec != NULL || use_all)
                                break;
 
                                break;
 
-                       WARNING(path, base, "Unknown directory part");
+                       if (warnings)
+                               say(path, "Unknown directory part");
                        fts_set(f, ff, FTS_SKIP);
                        break;
                        fts_set(f, ff, FTS_SKIP);
                        break;
-               case (2):
+               case 2:
                        /*
                         * Possibly our architecture.
                         * If we're descending, keep tabs on it.
                         */
                        /*
                         * Possibly our architecture.
                         * If we're descending, keep tabs on it.
                         */
-                       arch = NULL;
-                       if (FTS_DP != ff->fts_info && NULL != dsec)
-                               arch = stradd(ff->fts_name);
+                       if (ff->fts_info != FTS_DP && dsec != NULL)
+                               arch = ff->fts_name;
+                       else
+                               arch = NULL;
                        break;
                default:
                        break;
                default:
-                       if (FTS_DP == ff->fts_info || use_all)
+                       if (ff->fts_info == FTS_DP || use_all)
                                break;
                                break;
-                       WARNING(path, base, "Extraneous directory part");
+                       if (warnings)
+                               say(path, "Extraneous directory part");
                        fts_set(f, ff, FTS_SKIP);
                        break;
                }
        }
 
        fts_close(f);
                        fts_set(f, ff, FTS_SKIP);
                        break;
                }
        }
 
        fts_close(f);
-       return(1);
+       return 1;
 }
 
 /*
 }
 
 /*
- * Add a file to the file vector.
+ * Add a file to the mlinks table.
  * Do not verify that it's a "valid" looking manpage (we'll do that
  * later).
  *
  * Try to infer the manual section, architecture, and page name from the
  * path, assuming it looks like
  *
  * Do not verify that it's a "valid" looking manpage (we'll do that
  * later).
  *
  * Try to infer the manual section, architecture, and page name from the
  * path, assuming it looks like
  *
- *   [./]man*[/<arch>]/<name>.<section> 
+ *   [./]man*[/<arch>]/<name>.<section>
  *   or
  *   [./]cat<section>[/<arch>]/<name>.0
  *
  *   or
  *   [./]cat<section>[/<arch>]/<name>.0
  *
- * Stuff this information directly into the "of" vector.
  * See treescan() for the fts(3) version of this.
  */
 static void
  * See treescan() for the fts(3) version of this.
  */
 static void
-filescan(const char *file, const char *base)
+filescan(const char *file)
 {
 {
-       const char      *sec, *arch, *name, *dsec;
-       char            *p, *start, *buf;
-       int              dform;
+       char             buf[PATH_MAX];
        struct stat      st;
        struct stat      st;
+       struct mlink    *mlink;
+       char            *p, *start;
 
        assert(use_all);
 
        if (0 == strncmp(file, "./", 2))
                file += 2;
 
 
        assert(use_all);
 
        if (0 == strncmp(file, "./", 2))
                file += 2;
 
-       if (-1 == stat(file, &st)) {
-               WARNING(file, base, "%s", strerror(errno));
+       /*
+        * We have to do lstat(2) before realpath(3) loses
+        * the information whether this is a symbolic link.
+        * We need to know that because for symbolic links,
+        * we want to use the orginal file name, while for
+        * regular files, we want to use the real path.
+        */
+       if (-1 == lstat(file, &st)) {
+               exitcode = (int)MANDOCLEVEL_BADARG;
+               say(file, "&lstat");
                return;
                return;
-       } else if ( ! (S_IFREG & st.st_mode)) {
-               WARNING(file, base, "Not a regular file");
+       } else if (0 == ((S_IFREG | S_IFLNK) & st.st_mode)) {
+               exitcode = (int)MANDOCLEVEL_BADARG;
+               say(file, "Not a regular file");
                return;
                return;
-       } else if (inocheck(&st)) {
-               WARNING(file, base, "Duplicate file");
+       }
+
+       /*
+        * We have to resolve the file name to the real path
+        * in any case for the base directory check.
+        */
+       if (NULL == realpath(file, buf)) {
+               exitcode = (int)MANDOCLEVEL_BADARG;
+               say(file, "&realpath");
                return;
        }
 
                return;
        }
 
-       buf = mandoc_strdup(file);
-       start = buf;
-       sec = arch = name = dsec = NULL;
-       dform = FORM_NONE;
+       if (OP_TEST == op)
+               start = buf;
+       else if (strstr(buf, basedir) == buf)
+               start = buf + strlen(basedir);
+#ifdef HOMEBREWDIR
+       else if (strstr(buf, HOMEBREWDIR) == buf)
+               start = buf;
+#endif
+       else {
+               exitcode = (int)MANDOCLEVEL_BADARG;
+               say("", "%s: outside base directory", buf);
+               return;
+       }
+
+       /*
+        * Now we are sure the file is inside our tree.
+        * If it is a symbolic link, ignore the real path
+        * and use the original name.
+        * This implies passing stuff like "cat1/../man1/foo.1"
+        * on the command line won't work.  So don't do that.
+        * Note the stat(2) can still fail if the link target
+        * doesn't exist.
+        */
+       if (S_IFLNK & st.st_mode) {
+               if (-1 == stat(buf, &st)) {
+                       exitcode = (int)MANDOCLEVEL_BADARG;
+                       say(file, "&stat");
+                       return;
+               }
+               if (strlcpy(buf, file, sizeof(buf)) >= sizeof(buf)) {
+                       say(file, "Filename too long");
+                       return;
+               }
+               start = buf;
+               if (OP_TEST != op && strstr(buf, basedir) == buf)
+                       start += strlen(basedir);
+       }
+
+       mlink = mandoc_calloc(1, sizeof(struct mlink));
+       mlink->dform = FORM_NONE;
+       if (strlcpy(mlink->file, start, sizeof(mlink->file)) >=
+           sizeof(mlink->file)) {
+               say(start, "Filename too long");
+               free(mlink);
+               return;
+       }
+
+       /*
+        * In test mode or when the original name is absolute
+        * but outside our tree, guess the base directory.
+        */
+
+       if (op == OP_TEST || (start == buf && *start == '/')) {
+               if (strncmp(buf, "man/", 4) == 0)
+                       start = buf + 4;
+               else if ((start = strstr(buf, "/man/")) != NULL)
+                       start += 5;
+               else
+                       start = buf;
+       }
 
        /*
         * First try to guess our directory structure.
 
        /*
         * First try to guess our directory structure.
@@ -711,19 +893,19 @@ filescan(const char *file, const char *base)
        if (NULL != (p = strchr(start, '/'))) {
                *p++ = '\0';
                if (0 == strncmp(start, "man", 3)) {
        if (NULL != (p = strchr(start, '/'))) {
                *p++ = '\0';
                if (0 == strncmp(start, "man", 3)) {
-                       dform = FORM_SRC;
-                       dsec = start + 3;
+                       mlink->dform = FORM_SRC;
+                       mlink->dsec = start + 3;
                } else if (0 == strncmp(start, "cat", 3)) {
                } else if (0 == strncmp(start, "cat", 3)) {
-                       dform = FORM_CAT;
-                       dsec = start + 3;
+                       mlink->dform = FORM_CAT;
+                       mlink->dsec = start + 3;
                }
 
                start = p;
                }
 
                start = p;
-               if (NULL != dsec && NULL != (p = strchr(start, '/'))) {
+               if (NULL != mlink->dsec && NULL != (p = strchr(start, '/'))) {
                        *p++ = '\0';
                        *p++ = '\0';
-                       arch = start;
+                       mlink->arch = start;
                        start = p;
                        start = p;
-               } 
+               }
        }
 
        /*
        }
 
        /*
@@ -736,299 +918,387 @@ filescan(const char *file, const char *base)
 
        if ('.' == *p) {
                *p++ = '\0';
 
        if ('.' == *p) {
                *p++ = '\0';
-               sec = p;
+               mlink->fsec = p;
        }
 
        /*
         * Now try to parse the name.
         * Use the filename portion of the path.
         */
        }
 
        /*
         * Now try to parse the name.
         * Use the filename portion of the path.
         */
-       name = start;
+       mlink->name = start;
        if (NULL != (p = strrchr(start, '/'))) {
        if (NULL != (p = strrchr(start, '/'))) {
-               name = p + 1;
+               mlink->name = p + 1;
                *p = '\0';
                *p = '\0';
-       } 
-
-       ofadd(base, dform, file, name, dsec, sec, arch, &st);
-       free(buf);
+       }
+       mlink_add(mlink, &st);
 }
 
 }
 
-/*
- * See fileadd(). 
- */
-static int
-filecheck(const char *name)
+static void
+mlink_add(struct mlink *mlink, const struct stat *st)
 {
 {
-       unsigned int     index;
-
-       index = ohash_qlookup(&filenames, name);
-       return(NULL != ohash_find(&filenames, index));
+       struct inodev    inodev;
+       struct mpage    *mpage;
+       unsigned int     slot;
+
+       assert(NULL != mlink->file);
+
+       mlink->dsec = mandoc_strdup(mlink->dsec ? mlink->dsec : "");
+       mlink->arch = mandoc_strdup(mlink->arch ? mlink->arch : "");
+       mlink->name = mandoc_strdup(mlink->name ? mlink->name : "");
+       mlink->fsec = mandoc_strdup(mlink->fsec ? mlink->fsec : "");
+
+       if ('0' == *mlink->fsec) {
+               free(mlink->fsec);
+               mlink->fsec = mandoc_strdup(mlink->dsec);
+               mlink->fform = FORM_CAT;
+       } else if ('1' <= *mlink->fsec && '9' >= *mlink->fsec)
+               mlink->fform = FORM_SRC;
+       else
+               mlink->fform = FORM_NONE;
+
+       slot = ohash_qlookup(&mlinks, mlink->file);
+       assert(NULL == ohash_find(&mlinks, slot));
+       ohash_insert(&mlinks, slot, mlink);
+
+       memset(&inodev, 0, sizeof(inodev));  /* Clear padding. */
+       inodev.st_ino = st->st_ino;
+       inodev.st_dev = st->st_dev;
+       slot = ohash_lookup_memory(&mpages, (char *)&inodev,
+           sizeof(struct inodev), inodev.st_ino);
+       mpage = ohash_find(&mpages, slot);
+       if (NULL == mpage) {
+               mpage = mandoc_calloc(1, sizeof(struct mpage));
+               mpage->inodev.st_ino = inodev.st_ino;
+               mpage->inodev.st_dev = inodev.st_dev;
+               mpage->form = FORM_NONE;
+               mpage->next = mpage_head;
+               mpage_head = mpage;
+               ohash_insert(&mpages, slot, mpage);
+       } else
+               mlink->next = mpage->mlinks;
+       mpage->mlinks = mlink;
+       mlink->mpage = mpage;
 }
 
 }
 
-/*
- * Use the standard hashing mechanism (K&R) to see if the given filename
- * already exists.
- */
 static void
 static void
-fileadd(struct of *of)
+mlink_free(struct mlink *mlink)
 {
 {
-       unsigned int     index;
 
 
-       index = ohash_qlookup(&filenames, of->file);
-       assert(NULL == ohash_find(&filenames, index));
-       ohash_insert(&filenames, index, of);
+       free(mlink->dsec);
+       free(mlink->arch);
+       free(mlink->name);
+       free(mlink->fsec);
+       free(mlink);
 }
 
 }
 
-/*
- * See inoadd().
- */
-static int
-inocheck(const struct stat *st)
+static void
+mpages_free(void)
 {
 {
-       struct id        id;
-       uint32_t         hash;
-       unsigned int     index;
-
-       memset(&id, 0, sizeof(id));
-       id.ino = hash = st->st_ino;
-       id.dev = st->st_dev;
-       index = ohash_lookup_memory
-               (&inos, (char *)&id, sizeof(id), hash);
+       struct mpage    *mpage;
+       struct mlink    *mlink;
 
 
-       return(NULL != ohash_find(&inos, index));
+       while ((mpage = mpage_head) != NULL) {
+               while ((mlink = mpage->mlinks) != NULL) {
+                       mpage->mlinks = mlink->next;
+                       mlink_free(mlink);
+               }
+               mpage_head = mpage->next;
+               free(mpage->sec);
+               free(mpage->arch);
+               free(mpage->title);
+               free(mpage->desc);
+               free(mpage);
+       }
 }
 
 /*
 }
 
 /*
- * The hashing function used here is quite simple: simply take the inode
- * and use uint32_t of its bits.
- * Then when we do the lookup, use both the inode and device identifier.
+ * For each mlink to the mpage, check whether the path looks like
+ * it is formatted, and if it does, check whether a source manual
+ * exists by the same name, ignoring the suffix.
+ * If both conditions hold, drop the mlink.
  */
 static void
  */
 static void
-inoadd(const struct stat *st, struct of *of)
+mlinks_undupe(struct mpage *mpage)
 {
 {
-       uint32_t         hash;
-       unsigned int     index;
-
-       of->id.ino = hash = st->st_ino;
-       of->id.dev = st->st_dev;
-       index = ohash_lookup_memory
-               (&inos, (char *)&of->id, sizeof(of->id), hash);
-
-       assert(NULL == ohash_find(&inos, index));
-       ohash_insert(&inos, index, of);
+       char              buf[PATH_MAX];
+       struct mlink    **prev;
+       struct mlink     *mlink;
+       char             *bufp;
+
+       mpage->form = FORM_CAT;
+       prev = &mpage->mlinks;
+       while (NULL != (mlink = *prev)) {
+               if (FORM_CAT != mlink->dform) {
+                       mpage->form = FORM_NONE;
+                       goto nextlink;
+               }
+               (void)strlcpy(buf, mlink->file, sizeof(buf));
+               bufp = strstr(buf, "cat");
+               assert(NULL != bufp);
+               memcpy(bufp, "man", 3);
+               if (NULL != (bufp = strrchr(buf, '.')))
+                       *++bufp = '\0';
+               (void)strlcat(buf, mlink->dsec, sizeof(buf));
+               if (NULL == ohash_find(&mlinks,
+                   ohash_qlookup(&mlinks, buf)))
+                       goto nextlink;
+               if (warnings)
+                       say(mlink->file, "Man source exists: %s", buf);
+               if (use_all)
+                       goto nextlink;
+               *prev = mlink->next;
+               mlink_free(mlink);
+               continue;
+nextlink:
+               prev = &(*prev)->next;
+       }
 }
 
 static void
 }
 
 static void
-ofadd(const char *base, int dform, const char *file, 
-               const char *name, const char *dsec, const char *sec, 
-               const char *arch, const struct stat *st)
+mlink_check(struct mpage *mpage, struct mlink *mlink)
 {
 {
-       struct of       *of;
-       int              sform;
-
-       assert(NULL != file);
-
-       if (NULL == name)
-               name = "";
-       if (NULL == sec)
-               sec = "";
-       if (NULL == dsec)
-               dsec = "";
-       if (NULL == arch)
-               arch = "";
-
-       sform = FORM_NONE;
-       if (NULL != sec && *sec <= '9' && *sec >= '1')
-               sform = FORM_SRC;
-       else if (NULL != sec && *sec == '0') {
-               sec = dsec;
-               sform = FORM_CAT;
-       }
+       struct str      *str;
+       unsigned int     slot;
+
+       /*
+        * Check whether the manual section given in a file
+        * agrees with the directory where the file is located.
+        * Some manuals have suffixes like (3p) on their
+        * section number either inside the file or in the
+        * directory name, some are linked into more than one
+        * section, like encrypt(1) = makekey(8).
+        */
 
 
-       of = mandoc_calloc(1, sizeof(struct of));
-       strlcpy(of->file, file, MAXPATHLEN);
-       of->name = name;
-       of->sec = sec;
-       of->dsec = dsec;
-       of->arch = arch;
-       of->sform = sform;
-       of->dform = dform;
-       of->next = ofs;
-       ofs = of;
+       if (FORM_SRC == mpage->form &&
+           strcasecmp(mpage->sec, mlink->dsec))
+               say(mlink->file, "Section \"%s\" manual in %s directory",
+                   mpage->sec, mlink->dsec);
 
        /*
 
        /*
-        * Add to unique identifier hash.
-        * Then if it's a source manual and we're going to use source in
-        * favour of catpages, add it to that hash.
+        * Manual page directories exist for each kernel
+        * architecture as returned by machine(1).
+        * However, many manuals only depend on the
+        * application architecture as returned by arch(1).
+        * For example, some (2/ARM) manuals are shared
+        * across the "armish" and "zaurus" kernel
+        * architectures.
+        * A few manuals are even shared across completely
+        * different architectures, for example fdformat(1)
+        * on amd64, i386, and sparc64.
         */
         */
-       inoadd(st, of);
-       fileadd(of);
-}
 
 
-static void
-offree(void)
-{
-       struct of       *of;
+       if (strcasecmp(mpage->arch, mlink->arch))
+               say(mlink->file, "Architecture \"%s\" manual in "
+                   "\"%s\" directory", mpage->arch, mlink->arch);
 
 
-       while (NULL != (of = ofs)) {
-               ofs = of->next;
-               free(of);
-       }
+       /*
+        * XXX
+        * parse_cat() doesn't set NAME_TITLE yet.
+        */
+
+       if (FORM_CAT == mpage->form)
+               return;
+
+       /*
+        * Check whether this mlink
+        * appears as a name in the NAME section.
+        */
+
+       slot = ohash_qlookup(&names, mlink->name);
+       str = ohash_find(&names, slot);
+       assert(NULL != str);
+       if ( ! (NAME_TITLE & str->mask))
+               say(mlink->file, "Name missing in NAME section");
 }
 
 /*
 }
 
 /*
- * Run through the files in the global vector "ofs" and add them to the
- * database specified in "base".
+ * Run through the files in the global vector "mpages"
+ * and add them to the database specified in "basedir".
  *
  * This handles the parsing scheme itself, using the cues of directory
  * and filename to determine whether the file is parsable or not.
  */
  *
  * This handles the parsing scheme itself, using the cues of directory
  * and filename to determine whether the file is parsable or not.
  */
-static int
-ofmerge(struct mchars *mc, struct mparse *mp, const char *base)
+static void
+mpages_merge(struct dba *dba, struct mparse *mp)
 {
 {
-       int              form;
-       size_t           sz;
-       struct mdoc     *mdoc;
-       struct man      *man;
-       char             buf[MAXPATHLEN];
-       char            *bufp;
-       const char      *msec, *march, *mtitle, *cp;
-       struct of       *of;
-       enum mandoclevel lvl;
-
-       for (of = ofs; NULL != of; of = of->next) {
-               /*
-                * If we're a catpage (as defined by our path), then see
-                * if a manpage exists by the same name (ignoring the
-                * suffix).
-                * If it does, then we want to use it instead of our
-                * own.
-                */
-               if ( ! use_all && FORM_CAT == of->dform) {
-                       sz = strlcpy(buf, of->file, MAXPATHLEN);
-                       if (sz >= MAXPATHLEN) {
-                               WARNING(of->file, base, 
-                                       "Filename too long");
-                               continue;
-                       }
-                       bufp = strstr(buf, "cat");
-                       assert(NULL != bufp);
-                       memcpy(bufp, "man", 3);
-                       if (NULL != (bufp = strrchr(buf, '.')))
-                               *++bufp = '\0';
-                       strlcat(buf, of->dsec, MAXPATHLEN);
-                       if (filecheck(buf)) {
-                               WARNING(of->file, base, "Man "
-                                       "source exists: %s", buf);
-                               continue;
-                       }
-               }
+       struct mpage            *mpage, *mpage_dest;
+       struct mlink            *mlink, *mlink_dest;
+       struct roff_man         *man;
+       char                    *sodest;
+       char                    *cp;
+       int                      fd;
+
+       for (mpage = mpage_head; mpage != NULL; mpage = mpage->next) {
+               mlinks_undupe(mpage);
+               if ((mlink = mpage->mlinks) == NULL)
+                       continue;
 
 
-               words = NULL;
+               name_mask = NAME_MASK;
+               mandoc_ohash_init(&names, 4, offsetof(struct str, key));
+               mandoc_ohash_init(&strings, 6, offsetof(struct str, key));
                mparse_reset(mp);
                mparse_reset(mp);
-               mdoc = NULL;
                man = NULL;
                man = NULL;
-               form = 0;
-               msec = of->dsec;
-               march = of->arch;
-               mtitle = of->name;
+               sodest = NULL;
 
 
-               /*
-                * Try interpreting the file as mdoc(7) or man(7)
-                * source code, unless it is already known to be
-                * formatted.  Fall back to formatted mode.
-                */
-               if (FORM_SRC == of->dform || FORM_SRC == of->sform) {
-                       lvl = mparse_readfd(mp, -1, of->file);
-                       if (lvl < MANDOCLEVEL_FATAL)
-                               mparse_result(mp, &mdoc, &man);
-               } 
-
-               if (NULL != mdoc) {
-                       form = 1;
-                       msec = mdoc_meta(mdoc)->msec;
-                       march = mdoc_meta(mdoc)->arch;
-                       mtitle = mdoc_meta(mdoc)->title;
-               } else if (NULL != man) {
-                       form = 1;
-                       msec = man_meta(man)->msec;
-                       march = "";
-                       mtitle = man_meta(man)->title;
-               } 
-
-               if (NULL == msec) 
-                       msec = "";
-               if (NULL == march) 
-                       march = "";
-               if (NULL == mtitle) 
-                       mtitle = "";
+               if ((fd = mparse_open(mp, mlink->file)) == -1) {
+                       say(mlink->file, "&open");
+                       goto nextpage;
+               }
 
                /*
 
                /*
-                * Check whether the manual section given in a file
-                * agrees with the directory where the file is located.
-                * Some manuals have suffixes like (3p) on their
-                * section number either inside the file or in the
-                * directory name, some are linked into more than one
-                * section, like encrypt(1) = makekey(8).  Do not skip
-                * manuals for such reasons.
+                * Interpret the file as mdoc(7) or man(7) source
+                * code, unless it is known to be formatted.
                 */
                 */
-               if ( ! use_all && form && strcasecmp(msec, of->dsec))
-                       WARNING(of->file, base, "Section \"%s\" "
-                               "manual in %s directory", 
-                               msec, of->dsec);
+               if (mlink->dform != FORM_CAT || mlink->fform != FORM_CAT) {
+                       mparse_readfd(mp, fd, mlink->file);
+                       close(fd);
+                       fd = -1;
+                       mparse_result(mp, &man, &sodest);
+               }
 
 
-               /*
-                * Manual page directories exist for each kernel
-                * architecture as returned by machine(1).
-                * However, many manuals only depend on the
-                * application architecture as returned by arch(1).
-                * For example, some (2/ARM) manuals are shared
-                * across the "armish" and "zaurus" kernel
-                * architectures.
-                * A few manuals are even shared across completely
-                * different architectures, for example fdformat(1)
-                * on amd64, i386, sparc, and sparc64.
-                * Thus, warn about architecture mismatches,
-                * but don't skip manuals for this reason.
-                */
-               if ( ! use_all && strcasecmp(march, of->arch))
-                       WARNING(of->file, base, "Architecture \"%s\" "
-                               "manual in \"%s\" directory",
-                               march, of->arch);
-
-               putkey(of, of->name, TYPE_Nm);
-
-               if (NULL != mdoc) {
-                       if (NULL != (cp = mdoc_meta(mdoc)->name))
-                               putkey(of, cp, TYPE_Nm);
-                       parse_mdoc(of, mdoc_node(mdoc));
-               } else if (NULL != man)
-                       parse_man(of, man_node(man));
+               if (sodest != NULL) {
+                       mlink_dest = ohash_find(&mlinks,
+                           ohash_qlookup(&mlinks, sodest));
+                       if (mlink_dest == NULL) {
+                               mandoc_asprintf(&cp, "%s.gz", sodest);
+                               mlink_dest = ohash_find(&mlinks,
+                                   ohash_qlookup(&mlinks, cp));
+                               free(cp);
+                       }
+                       if (mlink_dest != NULL) {
+
+                               /* The .so target exists. */
+
+                               mpage_dest = mlink_dest->mpage;
+                               while (1) {
+                                       mlink->mpage = mpage_dest;
+
+                                       /*
+                                        * If the target was already
+                                        * processed, add the links
+                                        * to the database now.
+                                        * Otherwise, this will
+                                        * happen when we come
+                                        * to the target.
+                                        */
+
+                                       if (mpage_dest->dba != NULL)
+                                               dbadd_mlink(mlink);
+
+                                       if (mlink->next == NULL)
+                                               break;
+                                       mlink = mlink->next;
+                               }
+
+                               /* Move all links to the target. */
+
+                               mlink->next = mlink_dest->next;
+                               mlink_dest->next = mpage->mlinks;
+                               mpage->mlinks = NULL;
+                       }
+                       goto nextpage;
+               } else if (man != NULL && man->macroset == MACROSET_MDOC) {
+                       mdoc_validate(man);
+                       mpage->form = FORM_SRC;
+                       mpage->sec = man->meta.msec;
+                       mpage->sec = mandoc_strdup(
+                           mpage->sec == NULL ? "" : mpage->sec);
+                       mpage->arch = man->meta.arch;
+                       mpage->arch = mandoc_strdup(
+                           mpage->arch == NULL ? "" : mpage->arch);
+                       mpage->title = mandoc_strdup(man->meta.title);
+               } else if (man != NULL && man->macroset == MACROSET_MAN) {
+                       man_validate(man);
+                       if (*man->meta.msec != '\0' ||
+                           *man->meta.title != '\0') {
+                               mpage->form = FORM_SRC;
+                               mpage->sec = mandoc_strdup(man->meta.msec);
+                               mpage->arch = mandoc_strdup(mlink->arch);
+                               mpage->title = mandoc_strdup(man->meta.title);
+                       } else
+                               man = NULL;
+               }
+
+               assert(mpage->desc == NULL);
+               if (man == NULL) {
+                       mpage->form = FORM_CAT;
+                       mpage->sec = mandoc_strdup(mlink->dsec);
+                       mpage->arch = mandoc_strdup(mlink->arch);
+                       mpage->title = mandoc_strdup(mlink->name);
+                       parse_cat(mpage, fd);
+               } else if (man->macroset == MACROSET_MDOC)
+                       parse_mdoc(mpage, &man->meta, man->first);
                else
                else
-                       parse_catpage(of, base);
+                       parse_man(mpage, &man->meta, man->first);
+               if (mpage->desc == NULL) {
+                       mpage->desc = mandoc_strdup(mlink->name);
+                       if (warnings)
+                               say(mlink->file, "No one-line description, "
+                                   "using filename \"%s\"", mlink->name);
+               }
 
 
-               dbindex(mc, form, of, base);
-       }
+               for (mlink = mpage->mlinks;
+                    mlink != NULL;
+                    mlink = mlink->next) {
+                       putkey(mpage, mlink->name, NAME_FILE);
+                       if (warnings && !use_all)
+                               mlink_check(mpage, mlink);
+               }
 
 
-       return(1);
+               dbadd(dba, mpage);
+
+nextpage:
+               ohash_delete(&strings);
+               ohash_delete(&names);
+       }
 }
 
 static void
 }
 
 static void
-parse_catpage(struct of *of, const char *base)
+parse_cat(struct mpage *mpage, int fd)
 {
        FILE            *stream;
 {
        FILE            *stream;
-       char            *line, *p, *title;
-       size_t           len, plen, titlesz;
-
-       if (NULL == (stream = fopen(of->file, "r"))) {
-               WARNING(of->file, base, "%s", strerror(errno));
+       struct mlink    *mlink;
+       char            *line, *p, *title, *sec;
+       size_t           linesz, plen, titlesz;
+       ssize_t          len;
+       int              offs;
+
+       mlink = mpage->mlinks;
+       stream = fd == -1 ? fopen(mlink->file, "r") : fdopen(fd, "r");
+       if (stream == NULL) {
+               if (fd != -1)
+                       close(fd);
+               if (warnings)
+                       say(mlink->file, "&fopen");
                return;
        }
 
                return;
        }
 
+       line = NULL;
+       linesz = 0;
+
+       /* Parse the section number from the header line. */
+
+       while (getline(&line, &linesz, stream) != -1) {
+               if (*line == '\n')
+                       continue;
+               if ((sec = strchr(line, '(')) == NULL)
+                       break;
+               if ((p = strchr(++sec, ')')) == NULL)
+                       break;
+               free(mpage->sec);
+               mpage->sec = mandoc_strndup(sec, p - sec);
+               if (warnings && *mlink->dsec != '\0' &&
+                   strcasecmp(mpage->sec, mlink->dsec))
+                       say(mlink->file,
+                           "Section \"%s\" manual in %s directory",
+                           mpage->sec, mlink->dsec);
+               break;
+       }
+
        /* Skip to first blank line. */
 
        /* Skip to first blank line. */
 
-       while (NULL != (line = fgetln(stream, &len)))
-               if ('\n' == *line)
+       while (line == NULL || *line != '\n')
+               if (getline(&line, &linesz, stream) == -1)
                        break;
 
        /*
                        break;
 
        /*
@@ -1036,10 +1306,10 @@ parse_catpage(struct of *of, const char *base)
         * is the first section header.  Skip to it.
         */
 
         * is the first section header.  Skip to it.
         */
 
-       while (NULL != (line = fgetln(stream, &len)))
-               if ('\n' != *line && ' ' != *line)
+       while (getline(&line, &linesz, stream) != -1)
+               if (*line != '\n' && *line != ' ')
                        break;
                        break;
-       
+
        /*
         * Read up until the next section into a buffer.
         * Strip the leading and trailing newline from each read line,
        /*
         * Read up until the next section into a buffer.
         * Strip the leading and trailing newline from each read line,
@@ -1050,20 +1320,20 @@ parse_catpage(struct of *of, const char *base)
        titlesz = 0;
        title = NULL;
 
        titlesz = 0;
        title = NULL;
 
-       while (NULL != (line = fgetln(stream, &len))) {
-               if (' ' != *line || '\n' != line[len - 1])
+       while ((len = getline(&line, &linesz, stream)) != -1) {
+               if (*line != ' ')
                        break;
                        break;
-               while (len > 0 && isspace((unsigned char)*line)) {
-                       line++;
-                       len--;
-               }
-               if (1 == len)
+               offs = 0;
+               while (isspace((unsigned char)line[offs]))
+                       offs++;
+               if (line[offs] == '\0')
                        continue;
                        continue;
-               title = mandoc_realloc(title, titlesz + len);
-               memcpy(title + titlesz, line, len);
-               titlesz += len;
+               title = mandoc_realloc(title, titlesz + len - offs);
+               memcpy(title + titlesz, line + offs, len - offs);
+               titlesz += len - offs;
                title[titlesz - 1] = ' ';
        }
                title[titlesz - 1] = ' ';
        }
+       free(line);
 
        /*
         * If no page content can be found, or the input line
 
        /*
         * If no page content can be found, or the input line
@@ -1073,14 +1343,14 @@ parse_catpage(struct of *of, const char *base)
         */
 
        if (NULL == title || '\0' == *title) {
         */
 
        if (NULL == title || '\0' == *title) {
-               WARNING(of->file, base, "Cannot find NAME section");
+               if (warnings)
+                       say(mlink->file, "Cannot find NAME section");
                fclose(stream);
                free(title);
                return;
        }
 
                fclose(stream);
                free(title);
                return;
        }
 
-       title = mandoc_realloc(title, titlesz + 1);
-       title[titlesz] = '\0';
+       title[titlesz - 1] = '\0';
 
        /*
         * Skip to the first dash.
 
        /*
         * Skip to the first dash.
@@ -1092,7 +1362,9 @@ parse_catpage(struct of *of, const char *base)
                for (p += 2; ' ' == *p || '\b' == *p; p++)
                        /* Skip to next word. */ ;
        } else {
                for (p += 2; ' ' == *p || '\b' == *p; p++)
                        /* Skip to next word. */ ;
        } else {
-               WARNING(of->file, base, "No dash in title line");
+               if (warnings)
+                       say(mlink->file, "No dash in title line, "
+                           "reusing \"%s\" as one-line description", title);
                p = title;
        }
 
                p = title;
        }
 
@@ -1105,13 +1377,12 @@ parse_catpage(struct of *of, const char *base)
                if (0 == len) {
                        memmove(line, line + 1, plen--);
                        continue;
                if (0 == len) {
                        memmove(line, line + 1, plen--);
                        continue;
-               } 
+               }
                memmove(line - 1, line + 1, plen - len);
                plen -= 2;
        }
 
                memmove(line - 1, line + 1, plen - len);
                plen -= 2;
        }
 
-       of->desc = stradd(p);
-       putkey(of, p, TYPE_Nd);
+       mpage->desc = mandoc_strdup(p);
        fclose(stream);
        free(title);
 }
        fclose(stream);
        free(title);
 }
@@ -1120,48 +1391,40 @@ parse_catpage(struct of *of, const char *base)
  * Put a type/word pair into the word database for this particular file.
  */
 static void
  * Put a type/word pair into the word database for this particular file.
  */
 static void
-putkey(const struct of *of, const char *value, uint64_t type)
-{
-
-       assert(NULL != value);
-       wordaddbuf(of, value, strlen(value), type);
-}
-
-/*
- * Like putkey() but for unterminated strings.
- */
-static void
-putkeys(const struct of *of, const char *value, int sz, uint64_t type)
+putkey(const struct mpage *mpage, char *value, uint64_t type)
 {
 {
-
-       wordaddbuf(of, value, sz, type);
+       putkeys(mpage, value, strlen(value), type);
 }
 
 /*
  * Grok all nodes at or below a certain mdoc node into putkey().
  */
 static void
 }
 
 /*
  * Grok all nodes at or below a certain mdoc node into putkey().
  */
 static void
-putmdockey(const struct of *of, const struct mdoc_node *n, uint64_t m)
+putmdockey(const struct mpage *mpage,
+       const struct roff_node *n, uint64_t m, int taboo)
 {
 
        for ( ; NULL != n; n = n->next) {
 {
 
        for ( ; NULL != n; n = n->next) {
+               if (n->flags & taboo)
+                       continue;
                if (NULL != n->child)
                if (NULL != n->child)
-                       putmdockey(of, n->child, m);
-               if (MDOC_TEXT == n->type)
-                       putkey(of, n->string, m);
+                       putmdockey(mpage, n->child, m, taboo);
+               if (n->type == ROFFT_TEXT)
+                       putkey(mpage, n->string, m);
        }
 }
 
        }
 }
 
-static int
-parse_man(struct of *of, const struct man_node *n)
+static void
+parse_man(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 {
-       const struct man_node *head, *body;
-       char            *start, *sv, *title;
+       const struct roff_node *head, *body;
+       char            *start, *title;
        char             byte;
        char             byte;
-       size_t           sz, titlesz;
+       size_t           sz;
 
 
-       if (NULL == n)
-               return(0);
+       if (n == NULL)
+               return;
 
        /*
         * We're only searching for one thing: the first text child in
 
        /*
         * We're only searching for one thing: the first text child in
@@ -1170,19 +1433,14 @@ parse_man(struct of *of, const struct man_node *n)
         * the correct section or not.
         */
 
         * the correct section or not.
         */
 
-       if (MAN_BODY == n->type && MAN_SH == n->tok) {
+       if (n->type == ROFFT_BODY && n->tok == MAN_SH) {
                body = n;
                body = n;
-               assert(body->parent);
-               if (NULL != (head = body->parent->head) &&
-                               1 == head->nchild &&
-                               NULL != (head = (head->child)) &&
-                               MAN_TEXT == head->type &&
-                               0 == strcmp(head->string, "NAME") &&
-                               NULL != (body = body->child) &&
-                               MAN_TEXT == body->type) {
-
-                       title = NULL;
-                       titlesz = 0;
+               if ((head = body->parent->head) != NULL &&
+                   (head = head->child) != NULL &&
+                   head->next == NULL &&
+                   head->type == ROFFT_TEXT &&
+                   strcmp(head->string, "NAME") == 0 &&
+                   body->child != NULL) {
 
                        /*
                         * Suck the entire NAME section into memory.
 
                        /*
                         * Suck the entire NAME section into memory.
@@ -1191,48 +1449,12 @@ parse_man(struct of *of, const struct man_node *n)
                         * NAME sections over many lines.
                         */
 
                         * NAME sections over many lines.
                         */
 
-                       for ( ; NULL != body; body = body->next) {
-                               if (MAN_TEXT != body->type)
-                                       break;
-                               if (0 == (sz = strlen(body->string)))
-                                       continue;
-                               title = mandoc_realloc
-                                       (title, titlesz + sz + 1);
-                               memcpy(title + titlesz, body->string, sz);
-                               titlesz += sz + 1;
-                               title[titlesz - 1] = ' ';
-                       }
+                       title = NULL;
+                       deroff(&title, body);
                        if (NULL == title)
                        if (NULL == title)
-                               return(1);
-
-                       title = mandoc_realloc(title, titlesz + 1);
-                       title[titlesz] = '\0';
-
-                       /* Skip leading space.  */
-
-                       sv = title;
-                       while (isspace((unsigned char)*sv))
-                               sv++;
-
-                       if (0 == (sz = strlen(sv))) {
-                               free(title);
-                               return(1);
-                       }
-
-                       /* Erase trailing space. */
+                               return;
 
 
-                       start = &sv[sz - 1];
-                       while (start > sv && isspace((unsigned char)*start))
-                               *start-- = '\0';
-
-                       if (start == sv) {
-                               free(title);
-                               return(1);
-                       }
-
-                       start = sv;
-
-                       /* 
+                       /*
                         * Go through a special heuristic dance here.
                         * Conventionally, one or more manual names are
                         * comma-specified prior to a whitespace, then a
                         * Go through a special heuristic dance here.
                         * Conventionally, one or more manual names are
                         * comma-specified prior to a whitespace, then a
@@ -1240,6 +1462,7 @@ parse_man(struct of *of, const struct man_node *n)
                         * the name parts here.
                         */
 
                         * the name parts here.
                         */
 
+                       start = title;
                        for ( ;; ) {
                                sz = strcspn(start, " ,");
                                if ('\0' == start[sz])
                        for ( ;; ) {
                                sz = strcspn(start, " ,");
                                if ('\0' == start[sz])
@@ -1248,7 +1471,21 @@ parse_man(struct of *of, const struct man_node *n)
                                byte = start[sz];
                                start[sz] = '\0';
 
                                byte = start[sz];
                                start[sz] = '\0';
 
-                               putkey(of, start, TYPE_Nm);
+                               /*
+                                * Assume a stray trailing comma in the
+                                * name list if a name begins with a dash.
+                                */
+
+                               if ('-' == start[0] ||
+                                   ('\\' == start[0] && '-' == start[1]))
+                                       break;
+
+                               putkey(mpage, start, NAME_TITLE);
+                               if ( ! (mpage->name_head_done ||
+                                   strcasecmp(start, meta->title))) {
+                                       putkey(mpage, start, NAME_HEAD);
+                                       mpage->name_head_done = 1;
+                               }
 
                                if (' ' == byte) {
                                        start += sz + 1;
 
                                if (' ' == byte) {
                                        start += sz + 1;
@@ -1261,10 +1498,15 @@ parse_man(struct of *of, const struct man_node *n)
                                        start++;
                        }
 
                                        start++;
                        }
 
-                       if (sv == start) {
-                               putkey(of, start, TYPE_Nm);
+                       if (start == title) {
+                               putkey(mpage, start, NAME_TITLE);
+                               if ( ! (mpage->name_head_done ||
+                                   strcasecmp(start, meta->title))) {
+                                       putkey(mpage, start, NAME_HEAD);
+                                       mpage->name_head_done = 1;
+                               }
                                free(title);
                                free(title);
-                               return(1);
+                               return;
                        }
 
                        while (isspace((unsigned char)*start))
                        }
 
                        while (isspace((unsigned char)*start))
@@ -1284,63 +1526,62 @@ parse_man(struct of *of, const struct man_node *n)
                        while (' ' == *start)
                                start++;
 
                        while (' ' == *start)
                                start++;
 
-                       assert(NULL == of->desc);
-                       of->desc = stradd(start);
-                       putkey(of, start, TYPE_Nd);
+                       mpage->desc = mandoc_strdup(start);
                        free(title);
                        free(title);
-                       return(1);
+                       return;
                }
        }
 
                }
        }
 
-       for (n = n->child; n; n = n->next)
-               if (parse_man(of, n))
-                       return(1);
-
-       return(0);
+       for (n = n->child; n; n = n->next) {
+               if (NULL != mpage->desc)
+                       break;
+               parse_man(mpage, meta, n);
+       }
 }
 
 static void
 }
 
 static void
-parse_mdoc(struct of *of, const struct mdoc_node *n)
+parse_mdoc(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 
 {
 
-       assert(NULL != n);
-       for (n = n->child; NULL != n; n = n->next) {
+       for (n = n->child; n != NULL; n = n->next) {
+               if (n->tok == TOKEN_NONE ||
+                   n->tok < ROFF_MAX ||
+                   n->flags & mdocs[n->tok].taboo)
+                       continue;
+               assert(n->tok >= MDOC_Dd && n->tok < MDOC_MAX);
                switch (n->type) {
                switch (n->type) {
-               case (MDOC_ELEM):
-                       /* FALLTHROUGH */
-               case (MDOC_BLOCK):
-                       /* FALLTHROUGH */
-               case (MDOC_HEAD):
-                       /* FALLTHROUGH */
-               case (MDOC_BODY):
-                       /* FALLTHROUGH */
-               case (MDOC_TAIL):
-                       if (NULL != mdocs[n->tok].fp)
-                              if (0 == (*mdocs[n->tok].fp)(of, n))
-                                      break;
-
-                       if (MDOCF_CHILD & mdocs[n->tok].flags)
-                               putmdockey(of, n->child, mdocs[n->tok].mask);
+               case ROFFT_ELEM:
+               case ROFFT_BLOCK:
+               case ROFFT_HEAD:
+               case ROFFT_BODY:
+               case ROFFT_TAIL:
+                       if (mdocs[n->tok].fp != NULL &&
+                           (*mdocs[n->tok].fp)(mpage, meta, n) == 0)
+                               break;
+                       if (mdocs[n->tok].mask)
+                               putmdockey(mpage, n->child,
+                                   mdocs[n->tok].mask, mdocs[n->tok].taboo);
                        break;
                default:
                        break;
                default:
-                       assert(MDOC_ROOT != n->type);
                        continue;
                }
                if (NULL != n->child)
                        continue;
                }
                if (NULL != n->child)
-                       parse_mdoc(of, n);
+                       parse_mdoc(mpage, meta, n);
        }
 }
 
 static int
        }
 }
 
 static int
-parse_mdoc_Fd(struct of *of, const struct mdoc_node *n)
+parse_mdoc_Fd(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 {
-       const char      *start, *end;
+       char            *start, *end;
        size_t           sz;
 
        if (SEC_SYNOPSIS != n->sec ||
        size_t           sz;
 
        if (SEC_SYNOPSIS != n->sec ||
-                       NULL == (n = n->child) || 
-                       MDOC_TEXT != n->type)
-               return(0);
+           NULL == (n = n->child) ||
+           n->type != ROFFT_TEXT)
+               return 0;
 
        /*
         * Only consider those `Fd' macro fields that begin with an
 
        /*
         * Only consider those `Fd' macro fields that begin with an
@@ -1348,10 +1589,10 @@ parse_mdoc_Fd(struct of *of, const struct mdoc_node *n)
         */
 
        if (strcmp("#include", n->string))
         */
 
        if (strcmp("#include", n->string))
-               return(0);
+               return 0;
 
 
-       if (NULL == (n = n->next) || MDOC_TEXT != n->type)
-               return(0);
+       if ((n = n->next) == NULL || n->type != ROFFT_TEXT)
+               return 0;
 
        /*
         * Strip away the enclosing angle brackets and make sure we're
 
        /*
         * Strip away the enclosing angle brackets and make sure we're
@@ -1363,240 +1604,217 @@ parse_mdoc_Fd(struct of *of, const struct mdoc_node *n)
                start++;
 
        if (0 == (sz = strlen(start)))
                start++;
 
        if (0 == (sz = strlen(start)))
-               return(0);
+               return 0;
 
        end = &start[(int)sz - 1];
        if ('>' == *end || '"' == *end)
                end--;
 
        if (end > start)
 
        end = &start[(int)sz - 1];
        if ('>' == *end || '"' == *end)
                end--;
 
        if (end > start)
-               putkeys(of, start, end - start + 1, TYPE_In);
-       return(1);
+               putkeys(mpage, start, end - start + 1, TYPE_In);
+       return 0;
 }
 
 }
 
-static int
-parse_mdoc_In(struct of *of, const struct mdoc_node *n)
-{
-
-       if (NULL != n->child && MDOC_TEXT == n->child->type)
-               return(0);
-
-       putkey(of, n->child->string, TYPE_In);
-       return(1);
-}
-
-static int
-parse_mdoc_Fn(struct of *of, const struct mdoc_node *n)
+static void
+parse_mdoc_fname(struct mpage *mpage, const struct roff_node *n)
 {
 {
-       const char      *cp;
-
-       if (NULL == (n = n->child) || MDOC_TEXT != n->type)
-               return(0);
+       char    *cp;
+       size_t   sz;
 
 
-       /* 
-        * Parse: .Fn "struct type *name" "char *arg".
-        * First strip away pointer symbol. 
-        * Then store the function name, then type.
-        * Finally, store the arguments. 
-        */
+       if (n->type != ROFFT_TEXT)
+               return;
 
 
-       if (NULL == (cp = strrchr(n->string, ' ')))
-               cp = n->string;
+       /* Skip function pointer punctuation. */
 
 
-       while ('*' == *cp)
+       cp = n->string;
+       while (*cp == '(' || *cp == '*')
                cp++;
                cp++;
+       sz = strcspn(cp, "()");
 
 
-       putkey(of, cp, TYPE_Fn);
-
-       if (n->string < cp)
-               putkeys(of, n->string, cp - n->string, TYPE_Ft);
-
-       for (n = n->next; NULL != n; n = n->next)
-               if (MDOC_TEXT == n->type)
-                       putkey(of, n->string, TYPE_Fa);
-
-       return(0);
+       putkeys(mpage, cp, sz, TYPE_Fn);
+       if (n->sec == SEC_SYNOPSIS)
+               putkeys(mpage, cp, sz, NAME_SYN);
 }
 
 static int
 }
 
 static int
-parse_mdoc_St(struct of *of, const struct mdoc_node *n)
+parse_mdoc_Fn(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 
 {
 
-       if (NULL == n->child || MDOC_TEXT != n->child->type)
-               return(0);
+       if (n->child == NULL)
+               return 0;
 
 
-       putkey(of, n->child->string, TYPE_St);
-       return(1);
-}
+       parse_mdoc_fname(mpage, n->child);
 
 
-static int
-parse_mdoc_Xr(struct of *of, const struct mdoc_node *n)
-{
+       for (n = n->child->next; n != NULL; n = n->next)
+               if (n->type == ROFFT_TEXT)
+                       putkey(mpage, n->string, TYPE_Fa);
 
 
-       if (NULL == (n = n->child))
-               return(0);
-
-       putkey(of, n->string, TYPE_Xr);
-       return(1);
+       return 0;
 }
 
 static int
 }
 
 static int
-parse_mdoc_Nd(struct of *of, const struct mdoc_node *n)
+parse_mdoc_Fo(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 {
-       size_t           sz;
-       char            *sv, *desc;
-
-       if (MDOC_BODY != n->type)
-               return(0);
 
 
-       /*
-        * Special-case the `Nd' because we need to put the description
-        * into the document table.
-        */
+       if (n->type != ROFFT_HEAD)
+               return 1;
 
 
-       desc = NULL;
-       for (n = n->child; NULL != n; n = n->next) {
-               if (MDOC_TEXT == n->type) {
-                       sz = strlen(n->string) + 1;
-                       if (NULL != (sv = desc))
-                               sz += strlen(desc) + 1;
-                       desc = mandoc_realloc(desc, sz);
-                       if (NULL != sv)
-                               strlcat(desc, " ", sz);
-                       else
-                               *desc = '\0';
-                       strlcat(desc, n->string, sz);
-               }
-               if (NULL != n->child)
-                       parse_mdoc_Nd(of, n);
-       }
+       if (n->child != NULL)
+               parse_mdoc_fname(mpage, n->child);
 
 
-       of->desc = NULL != desc ? stradd(desc) : NULL;
-       free(desc);
-       return(1);
+       return 0;
 }
 
 static int
 }
 
 static int
-parse_mdoc_Nm(struct of *of, const struct mdoc_node *n)
+parse_mdoc_Va(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 {
+       char *cp;
+
+       if (n->type != ROFFT_ELEM && n->type != ROFFT_BODY)
+               return 0;
+
+       if (n->child != NULL &&
+           n->child->next == NULL &&
+           n->child->type == ROFFT_TEXT)
+               return 1;
+
+       cp = NULL;
+       deroff(&cp, n);
+       if (cp != NULL) {
+               putkey(mpage, cp, TYPE_Vt | (n->tok == MDOC_Va ||
+                   n->type == ROFFT_BODY ? TYPE_Va : 0));
+               free(cp);
+       }
 
 
-       if (SEC_NAME == n->sec)
-               return(1);
-       else if (SEC_SYNOPSIS != n->sec || MDOC_HEAD != n->type)
-               return(0);
-
-       return(1);
+       return 0;
 }
 
 static int
 }
 
 static int
-parse_mdoc_Sh(struct of *of, const struct mdoc_node *n)
+parse_mdoc_Xr(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 {
+       char    *cp;
 
 
-       return(SEC_CUSTOM == n->sec && MDOC_HEAD == n->type);
-}
+       if (NULL == (n = n->child))
+               return 0;
 
 
-static int
-parse_mdoc_head(struct of *of, const struct mdoc_node *n)
-{
+       if (NULL == n->next) {
+               putkey(mpage, n->string, TYPE_Xr);
+               return 0;
+       }
 
 
-       return(MDOC_HEAD == n->type);
+       mandoc_asprintf(&cp, "%s(%s)", n->string, n->next->string);
+       putkey(mpage, cp, TYPE_Xr);
+       free(cp);
+       return 0;
 }
 
 static int
 }
 
 static int
-parse_mdoc_body(struct of *of, const struct mdoc_node *n)
+parse_mdoc_Nd(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 
 {
 
-       return(MDOC_BODY == n->type);
+       if (n->type == ROFFT_BODY)
+               deroff(&mpage->desc, n);
+       return 0;
 }
 
 }
 
-/*
- * See straddbuf().
- */
-static char *
-stradd(const char *cp)
+static int
+parse_mdoc_Nm(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 
 {
 
-       return(straddbuf(cp, strlen(cp)));
+       if (SEC_NAME == n->sec)
+               putmdockey(mpage, n->child, NAME_TITLE, 0);
+       else if (n->sec == SEC_SYNOPSIS && n->type == ROFFT_HEAD) {
+               if (n->child == NULL)
+                       putkey(mpage, meta->name, NAME_SYN);
+               else
+                       putmdockey(mpage, n->child, NAME_SYN, 0);
+       }
+       if ( ! (mpage->name_head_done ||
+           n->child == NULL || n->child->string == NULL ||
+           strcasecmp(n->child->string, meta->title))) {
+               putkey(mpage, n->child->string, NAME_HEAD);
+               mpage->name_head_done = 1;
+       }
+       return 0;
 }
 
 }
 
-/*
- * This looks up or adds a string to the string table.
- * The string table is a table of all strings encountered during parse
- * or file scan.
- * In using it, we avoid having thousands of (e.g.) "cat1" string
- * allocations for the "of" table.
- * We also have a layer atop the string table for keeping track of words
- * in a parse sequence (see wordaddbuf()).
- */
-static char *
-straddbuf(const char *cp, size_t sz)
+static int
+parse_mdoc_Sh(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 {
-       struct str      *s;
-       unsigned int     index;
-       const char      *end;
-
-       if (NULL != (s = hashget(cp, sz)))
-               return(s->key);
-
-       s = mandoc_calloc(sizeof(struct str) + sz, 1);
-       memcpy(s->key, cp, sz);
 
 
-       end = cp + sz;
-       index = ohash_qlookupi(&strings, cp, &end);
-       assert(NULL == ohash_find(&strings, index));
-       ohash_insert(&strings, index, s);
-       return(s->key);
+       return n->sec == SEC_CUSTOM && n->type == ROFFT_HEAD;
 }
 
 }
 
-static struct str *
-hashget(const char *cp, size_t sz)
+static int
+parse_mdoc_head(struct mpage *mpage, const struct roff_meta *meta,
+       const struct roff_node *n)
 {
 {
-       unsigned int     index;
-       const char      *end;
 
 
-       end = cp + sz;
-       index = ohash_qlookupi(&strings, cp, &end);
-       return(ohash_find(&strings, index));
+       return n->type == ROFFT_HEAD;
 }
 
 /*
 }
 
 /*
- * Add a word to the current parse sequence.
- * Within the hashtable of strings, we maintain a list of strings that
- * are currently indexed.
- * Each of these ("words") has a bitmask modified within the parse.
- * When we finish a parse, we'll dump the list, then remove the head
- * entry -- since the next parse will have a new "of", it can keep track
- * of its entries without conflict.
+ * Add a string to the hash table for the current manual.
+ * Each string has a bitmask telling which macros it belongs to.
+ * When we finish the manual, we'll dump the table.
  */
 static void
  */
 static void
-wordaddbuf(const struct of *of, 
-               const char *cp, size_t sz, uint64_t v)
+putkeys(const struct mpage *mpage, char *cp, size_t sz, uint64_t v)
 {
 {
+       struct ohash    *htab;
        struct str      *s;
        struct str      *s;
-       unsigned int     index;
        const char      *end;
        const char      *end;
+       unsigned int     slot;
+       int              i, mustfree;
 
        if (0 == sz)
                return;
 
 
        if (0 == sz)
                return;
 
-       s = hashget(cp, sz);
+       mustfree = render_string(&cp, &sz);
+
+       if (TYPE_Nm & v) {
+               htab = &names;
+               v &= name_mask;
+               if (v & NAME_FIRST)
+                       name_mask &= ~NAME_FIRST;
+               if (debug > 1)
+                       say(mpage->mlinks->file,
+                           "Adding name %*s, bits=0x%llx", (int)sz, cp,
+                           (unsigned long long)v);
+       } else {
+               htab = &strings;
+               if (debug > 1)
+                   for (i = 0; i < KEY_MAX; i++)
+                       if ((uint64_t)1 << i & v)
+                           say(mpage->mlinks->file,
+                               "Adding key %s=%*s",
+                               mansearch_keynames[i], (int)sz, cp);
+       }
+
+       end = cp + sz;
+       slot = ohash_qlookupi(htab, cp, &end);
+       s = ohash_find(htab, slot);
 
 
-       if (NULL != s && of == s->of) {
+       if (NULL != s && mpage == s->mpage) {
                s->mask |= v;
                return;
        } else if (NULL == s) {
                s->mask |= v;
                return;
        } else if (NULL == s) {
-               s = mandoc_calloc(sizeof(struct str) + sz, 1);
+               s = mandoc_calloc(1, sizeof(struct str) + sz + 1);
                memcpy(s->key, cp, sz);
                memcpy(s->key, cp, sz);
-               end = cp + sz;
-               index = ohash_qlookupi(&strings, cp, &end);
-               assert(NULL == ohash_find(&strings, index));
-               ohash_insert(&strings, index, s);
+               ohash_insert(htab, slot, s);
        }
        }
-
-       s->next = words;
-       s->of = of;
+       s->mpage = mpage;
        s->mask = v;
        s->mask = v;
-       words = s;
+
+       if (mustfree)
+               free(cp);
 }
 
 /*
 }
 
 /*
@@ -1645,347 +1863,476 @@ utf8(unsigned int cp, char out[7])
                out[4] = (cp >> 6  & 63) | 128;
                out[5] = (cp       & 63) | 128;
        } else
                out[4] = (cp >> 6  & 63) | 128;
                out[5] = (cp       & 63) | 128;
        } else
-               return(0);
+               return 0;
 
        out[rc] = '\0';
 
        out[rc] = '\0';
-       return(rc);
+       return rc;
 }
 
 /*
 }
 
 /*
- * Store the UTF-8 version of a key, or alias the pointer if the key has
- * no UTF-8 transcription marks in it.
+ * If the string contains escape sequences,
+ * replace it with an allocated rendering and return 1,
+ * such that the caller can free it after use.
+ * Otherwise, do nothing and return 0.
  */
  */
-static void
-utf8key(struct mchars *mc, struct str *key)
+static int
+render_string(char **public, size_t *psz)
 {
 {
-       size_t           sz, bsz, pos;
-       char             utfbuf[7], res[5];
-       char            *buf;
-       const char      *seq, *cpp, *val;
-       int              len, u;
-       enum mandoc_esc  esc;
-
-       assert(NULL == key->utf8);
+       const char      *src, *scp, *addcp, *seq;
+       char            *dst;
+       size_t           ssz, dsz, addsz;
+       char             utfbuf[7], res[6];
+       int              seqlen, unicode;
 
        res[0] = '\\';
        res[1] = '\t';
        res[2] = ASCII_NBRSP;
        res[3] = ASCII_HYPH;
 
        res[0] = '\\';
        res[1] = '\t';
        res[2] = ASCII_NBRSP;
        res[3] = ASCII_HYPH;
-       res[4] = '\0';
+       res[4] = ASCII_BREAK;
+       res[5] = '\0';
 
 
-       val = key->key;
-       bsz = strlen(val);
+       src = scp = *public;
+       ssz = *psz;
+       dst = NULL;
+       dsz = 0;
 
 
-       /*
-        * Pre-check: if we have no stop-characters, then set the
-        * pointer as ourselvse and get out of here.
-        */
-       if (strcspn(val, res) == bsz) {
-               key->utf8 = key->key;
-               return;
-       } 
+       while (scp < src + *psz) {
 
 
-       /* Pre-allocate by the length of the input */
+               /* Leave normal characters unchanged. */
 
 
-       buf = mandoc_malloc(++bsz);
-       pos = 0;
+               if (strchr(res, *scp) == NULL) {
+                       if (dst != NULL)
+                               dst[dsz++] = *scp;
+                       scp++;
+                       continue;
+               }
 
 
-       while ('\0' != *val) {
                /*
                /*
-                * Halt on the first escape sequence.
-                * This also halts on the end of string, in which case
-                * we just copy, fallthrough, and exit the loop.
+                * Found something that requires replacing,
+                * make sure we have a destination buffer.
                 */
                 */
-               if ((sz = strcspn(val, res)) > 0) {
-                       memcpy(&buf[pos], val, sz);
-                       pos += sz;
-                       val += sz;
+
+               if (dst == NULL) {
+                       dst = mandoc_malloc(ssz + 1);
+                       dsz = scp - src;
+                       memcpy(dst, src, dsz);
                }
 
                }
 
-               if (ASCII_HYPH == *val) {
-                       buf[pos++] = '-';
-                       val++;
+               /* Handle single-char special characters. */
+
+               switch (*scp) {
+               case '\\':
+                       break;
+               case '\t':
+               case ASCII_NBRSP:
+                       dst[dsz++] = ' ';
+                       scp++;
                        continue;
                        continue;
-               } else if ('\t' == *val || ASCII_NBRSP == *val) {
-                       buf[pos++] = ' ';
-                       val++;
+               case ASCII_HYPH:
+                       dst[dsz++] = '-';
+                       /* FALLTHROUGH */
+               case ASCII_BREAK:
+                       scp++;
                        continue;
                        continue;
-               } else if ('\\' != *val)
-                       break;
-
-               /* Read past the slash. */
-
-               val++;
-               u = 0;
+               default:
+                       abort();
+               }
 
                /*
 
                /*
-                * Parse the escape sequence and see if it's a
-                * predefined character or special character.
+                * Found an escape sequence.
+                * Read past the slash, then parse it.
+                * Ignore everything except characters.
                 */
                 */
-               esc = mandoc_escape
-                       ((const char **)&val, &seq, &len);
-               if (ESCAPE_ERROR == esc)
-                       break;
 
 
-               if (ESCAPE_SPECIAL != esc)
-                       continue;
-               if (0 == (u = mchars_spec2cp(mc, seq, len)))
+               scp++;
+               if (mandoc_escape(&scp, &seq, &seqlen) != ESCAPE_SPECIAL)
                        continue;
 
                /*
                        continue;
 
                /*
-                * If we have a Unicode codepoint, try to convert that
-                * to a UTF-8 byte string.
+                * Render the special character
+                * as either UTF-8 or ASCII.
                 */
                 */
-               cpp = utfbuf;
-               if (0 == (sz = utf8(u, utfbuf)))
-                       continue;
+
+               if (write_utf8) {
+                       unicode = mchars_spec2cp(seq, seqlen);
+                       if (unicode <= 0)
+                               continue;
+                       addsz = utf8(unicode, utfbuf);
+                       if (addsz == 0)
+                               continue;
+                       addcp = utfbuf;
+               } else {
+                       addcp = mchars_spec2str(seq, seqlen, &addsz);
+                       if (addcp == NULL)
+                               continue;
+                       if (*addcp == ASCII_NBRSP) {
+                               addcp = " ";
+                               addsz = 1;
+                       }
+               }
 
                /* Copy the rendered glyph into the stream. */
 
 
                /* Copy the rendered glyph into the stream. */
 
-               sz = strlen(cpp);
-               bsz += sz;
+               ssz += addsz;
+               dst = mandoc_realloc(dst, ssz + 1);
+               memcpy(dst + dsz, addcp, addsz);
+               dsz += addsz;
+       }
+       if (dst != NULL) {
+               *public = dst;
+               *psz = dsz;
+       }
 
 
-               buf = mandoc_realloc(buf, bsz);
+       /* Trim trailing whitespace and NUL-terminate. */
 
 
-               memcpy(&buf[pos], cpp, sz);
-               pos += sz;
-       }
+       while (*psz > 0 && (*public)[*psz - 1] == ' ')
+               --*psz;
+       if (dst != NULL) {
+               (*public)[*psz] = '\0';
+               return 1;
+       } else
+               return 0;
+}
 
 
-       buf[pos] = '\0';
-       key->utf8 = buf;
+static void
+dbadd_mlink(const struct mlink *mlink)
+{
+       dba_page_alias(mlink->mpage->dba, mlink->name, NAME_FILE);
+       dba_page_add(mlink->mpage->dba, DBP_SECT, mlink->dsec);
+       dba_page_add(mlink->mpage->dba, DBP_SECT, mlink->fsec);
+       dba_page_add(mlink->mpage->dba, DBP_ARCH, mlink->arch);
+       dba_page_add(mlink->mpage->dba, DBP_FILE, mlink->file);
 }
 
 /*
  * Flush the current page's terms (and their bits) into the database.
 }
 
 /*
  * Flush the current page's terms (and their bits) into the database.
- * Wrap the entire set of additions in a transaction to make sqlite be a
- * little faster.
- * Also, UTF-8-encode the description at the last possible moment.
+ * Also, handle escape sequences at the last possible moment.
  */
 static void
  */
 static void
-dbindex(struct mchars *mc, int form, 
-               const struct of *of, const char *base)
+dbadd(struct dba *dba, struct mpage *mpage)
 {
 {
+       struct mlink    *mlink;
        struct str      *key;
        struct str      *key;
-       const char      *desc;
-       int64_t          recno;
-
-       DEBUG(of->file, base, "Adding to index");
-
-       if (nodb)
+       char            *cp;
+       uint64_t         mask;
+       size_t           i;
+       unsigned int     slot;
+       int              mustfree;
+
+       mlink = mpage->mlinks;
+
+       if (nodb) {
+               for (key = ohash_first(&names, &slot); NULL != key;
+                    key = ohash_next(&names, &slot))
+                       free(key);
+               for (key = ohash_first(&strings, &slot); NULL != key;
+                    key = ohash_next(&strings, &slot))
+                       free(key);
+               if (0 == debug)
+                       return;
+               while (NULL != mlink) {
+                       fputs(mlink->name, stdout);
+                       if (NULL == mlink->next ||
+                           strcmp(mlink->dsec, mlink->next->dsec) ||
+                           strcmp(mlink->fsec, mlink->next->fsec) ||
+                           strcmp(mlink->arch, mlink->next->arch)) {
+                               putchar('(');
+                               if ('\0' == *mlink->dsec)
+                                       fputs(mlink->fsec, stdout);
+                               else
+                                       fputs(mlink->dsec, stdout);
+                               if ('\0' != *mlink->arch)
+                                       printf("/%s", mlink->arch);
+                               putchar(')');
+                       }
+                       mlink = mlink->next;
+                       if (NULL != mlink)
+                               fputs(", ", stdout);
+               }
+               printf(" - %s\n", mpage->desc);
                return;
                return;
-
-       desc = "";
-       if (NULL != of->desc) {
-               key = hashget(of->desc, strlen(of->desc));
-               assert(NULL != key);
-               if (NULL == key->utf8)
-                       utf8key(mc, key);
-               desc = key->utf8;
        }
 
        }
 
-       sqlite3_exec(db, "BEGIN TRANSACTION", NULL, NULL, NULL);
-
-       sqlite3_bind_text
-               (stmts[STMT_INSERT_DOC], 1, 
-                of->file, -1, SQLITE_STATIC);
-       sqlite3_bind_text
-               (stmts[STMT_INSERT_DOC], 2, 
-                of->sec, -1, SQLITE_STATIC);
-       sqlite3_bind_text
-               (stmts[STMT_INSERT_DOC], 3, 
-                of->arch, -1, SQLITE_STATIC);
-       sqlite3_bind_text
-               (stmts[STMT_INSERT_DOC], 4, 
-                desc, -1, SQLITE_STATIC);
-       sqlite3_bind_int
-               (stmts[STMT_INSERT_DOC], 5, form);
-       sqlite3_step(stmts[STMT_INSERT_DOC]);
-       recno = sqlite3_last_insert_rowid(db);
-       sqlite3_reset(stmts[STMT_INSERT_DOC]);
-
-       for (key = words; NULL != key; key = key->next) {
-               assert(key->of == of);
-               if (NULL == key->utf8)
-                       utf8key(mc, key);
-               sqlite3_bind_int64
-                       (stmts[STMT_INSERT_KEY], 1, key->mask);
-               sqlite3_bind_text
-                       (stmts[STMT_INSERT_KEY], 2, 
-                        key->utf8, -1, SQLITE_STATIC);
-               sqlite3_bind_int64
-                       (stmts[STMT_INSERT_KEY], 3, recno);
-               sqlite3_step(stmts[STMT_INSERT_KEY]);
-               sqlite3_reset(stmts[STMT_INSERT_KEY]);
+       if (debug)
+               say(mlink->file, "Adding to database");
+
+       cp = mpage->desc;
+       i = strlen(cp);
+       mustfree = render_string(&cp, &i);
+       mpage->dba = dba_page_new(dba->pages,
+           *mpage->arch == '\0' ? mlink->arch : mpage->arch,
+           cp, mlink->file, mpage->form);
+       if (mustfree)
+               free(cp);
+       dba_page_add(mpage->dba, DBP_SECT, mpage->sec);
+
+       while (mlink != NULL) {
+               dbadd_mlink(mlink);
+               mlink = mlink->next;
        }
 
        }
 
-       sqlite3_exec(db, "COMMIT TRANSACTION", NULL, NULL, NULL);
-
+       for (key = ohash_first(&names, &slot); NULL != key;
+            key = ohash_next(&names, &slot)) {
+               assert(key->mpage == mpage);
+               dba_page_alias(mpage->dba, key->key, key->mask);
+               free(key);
+       }
+       for (key = ohash_first(&strings, &slot); NULL != key;
+            key = ohash_next(&strings, &slot)) {
+               assert(key->mpage == mpage);
+               i = 0;
+               for (mask = TYPE_Xr; mask <= TYPE_Lb; mask *= 2) {
+                       if (key->mask & mask)
+                               dba_macro_add(dba->macros, i,
+                                   key->key, mpage->dba);
+                       i++;
+               }
+               free(key);
+       }
 }
 
 static void
 }
 
 static void
-dbprune(const char *base)
+dbprune(struct dba *dba)
 {
 {
-       struct of       *of;
-
-       if (nodb)
-               return;
-
-       for (of = ofs; NULL != of; of = of->next) {
-               sqlite3_bind_text
-                       (stmts[STMT_DELETE], 1, 
-                        of->file, -1, SQLITE_STATIC);
-               sqlite3_step(stmts[STMT_DELETE]);
-               sqlite3_reset(stmts[STMT_DELETE]);
-               DEBUG(of->file, base, "Deleted from index");
+       struct dba_array        *page, *files;
+       char                    *file;
+
+       dba_array_FOREACH(dba->pages, page) {
+               files = dba_array_get(page, DBP_FILE);
+               dba_array_FOREACH(files, file) {
+                       if (*file < ' ')
+                               file++;
+                       if (ohash_find(&mlinks, ohash_qlookup(&mlinks,
+                           file)) != NULL) {
+                               if (debug)
+                                       say(file, "Deleting from database");
+                               dba_array_del(dba->pages);
+                               break;
+                       }
+               }
        }
 }
 
 /*
        }
 }
 
 /*
- * Close an existing database and its prepared statements.
- * If "real" is not set, rename the temporary file into the real one.
+ * Write the database from memory to disk.
  */
 static void
  */
 static void
-dbclose(const char *base, int real)
+dbwrite(struct dba *dba)
 {
 {
-       size_t           i;
-       char             file[MAXPATHLEN];
+       char             tfn[32];
+       int              status;
+       pid_t            child;
 
 
-       if (nodb)
-               return;
+       /*
+        * Do not write empty databases, and delete existing ones
+        * when makewhatis -u causes them to become empty.
+        */
 
 
-       for (i = 0; i < STMT__MAX; i++) {
-               sqlite3_finalize(stmts[i]);
-               stmts[i] = NULL;
+       dba_array_start(dba->pages);
+       if (dba_array_next(dba->pages) == NULL) {
+               if (unlink(MANDOC_DB) == -1)
+                       say(MANDOC_DB, "&unlink");
+               return;
        }
 
        }
 
-       sqlite3_close(db);
-       db = NULL;
+       /*
+        * Build the database in a temporary file,
+        * then atomically move it into place.
+        */
 
 
-       if (real)
+       if (dba_write(MANDOC_DB "~", dba) != -1) {
+               if (rename(MANDOC_DB "~", MANDOC_DB) == -1) {
+                       exitcode = (int)MANDOCLEVEL_SYSERR;
+                       say(MANDOC_DB, "&rename");
+                       unlink(MANDOC_DB "~");
+               }
                return;
                return;
-
-       strlcpy(file, MANDOC_DB, MAXPATHLEN);
-       strlcat(file, "~", MAXPATHLEN);
-       if (-1 == rename(file, MANDOC_DB))
-               perror(MANDOC_DB);
-}
-
-/*
- * This is straightforward stuff.
- * Open a database connection to a "temporary" database, then open a set
- * of prepared statements we'll use over and over again.
- * If "real" is set, we use the existing database; if not, we truncate a
- * temporary one.
- * Must be matched by dbclose().
- */
-static int
-dbopen(const char *base, int real)
-{
-       char             file[MAXPATHLEN];
-       const char      *sql;
-       int              rc, ofl;
-       size_t           sz;
-
-       if (nodb) 
-               return(1);
-
-       sz = strlcpy(file, MANDOC_DB, MAXPATHLEN);
-       if ( ! real)
-               sz = strlcat(file, "~", MAXPATHLEN);
-
-       if (sz >= MAXPATHLEN) {
-               fprintf(stderr, "%s: Path too long\n", file);
-               return(0);
        }
 
        }
 
-       if ( ! real)
-               remove(file);
-
-       ofl = SQLITE_OPEN_PRIVATECACHE | SQLITE_OPEN_READWRITE |
-               (0 == real ? SQLITE_OPEN_EXCLUSIVE : 0);
+       /*
+        * We lack write permission and cannot replace the database
+        * file, but let's at least check whether the data changed.
+        */
 
 
-       rc = sqlite3_open_v2(file, &db, ofl, NULL);
-       if (SQLITE_OK == rc) 
-               return(1);
-       if (SQLITE_CANTOPEN != rc) {
-               perror(file);
-               return(0);
+       (void)strlcpy(tfn, "/tmp/mandocdb.XXXXXXXX", sizeof(tfn));
+       if (mkdtemp(tfn) == NULL) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "&%s", tfn);
+               return;
        }
 
        }
 
-       sqlite3_close(db);
-       db = NULL;
-
-       if (SQLITE_OK != (rc = sqlite3_open(file, &db))) {
-               perror(file);
-               return(0);
+       (void)strlcat(tfn, "/" MANDOC_DB, sizeof(tfn));
+       if (dba_write(tfn, dba) == -1) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say(tfn, "&dba_write");
+               goto out;
        }
 
        }
 
-       sql = "CREATE TABLE \"docs\" (\n"
-             " \"file\" TEXT NOT NULL,\n"
-             " \"sec\" TEXT NOT NULL,\n"
-             " \"arch\" TEXT NOT NULL,\n"
-             " \"desc\" TEXT NOT NULL,\n"
-             " \"form\" INTEGER NOT NULL,\n"
-             " \"id\" INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL\n"
-             ");\n"
-             "\n"
-             "CREATE TABLE \"keys\" (\n"
-             " \"bits\" INTEGER NOT NULL,\n"
-             " \"key\" TEXT NOT NULL,\n"
-             " \"docid\" INTEGER NOT NULL REFERENCES docs(id) "
-               "ON DELETE CASCADE,\n"
-             " \"id\" INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL\n"
-             ");\n"
-             "\n"
-             "CREATE INDEX \"key_index\" ON keys (key);\n";
-
-       if (SQLITE_OK != sqlite3_exec(db, sql, NULL, NULL, NULL)) {
-               perror(sqlite3_errmsg(db));
-               return(0);
+       switch (child = fork()) {
+       case -1:
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "&fork cmp");
+               return;
+       case 0:
+               execlp("cmp", "cmp", "-s", tfn, MANDOC_DB, (char *)NULL);
+               say("", "&exec cmp");
+               exit(0);
+       default:
+               break;
+       }
+       if (waitpid(child, &status, 0) == -1) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "&wait cmp");
+       } else if (WIFSIGNALED(status)) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "cmp died from signal %d", WTERMSIG(status));
+       } else if (WEXITSTATUS(status)) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say(MANDOC_DB,
+                   "Data changed, but cannot replace database");
        }
 
        }
 
-       sql = "DELETE FROM docs where file=?";
-       sqlite3_prepare_v2(db, sql, -1, &stmts[STMT_DELETE], NULL);
-       sql = "INSERT INTO docs "
-               "(file,sec,arch,desc,form) VALUES (?,?,?,?,?)";
-       sqlite3_prepare_v2(db, sql, -1, &stmts[STMT_INSERT_DOC], NULL);
-       sql = "INSERT INTO keys "
-               "(bits,key,docid) VALUES (?,?,?)";
-       sqlite3_prepare_v2(db, sql, -1, &stmts[STMT_INSERT_KEY], NULL);
-       return(1);
+out:
+       *strrchr(tfn, '/') = '\0';
+       switch (child = fork()) {
+       case -1:
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "&fork rm");
+               return;
+       case 0:
+               execlp("rm", "rm", "-rf", tfn, (char *)NULL);
+               say("", "&exec rm");
+               exit((int)MANDOCLEVEL_SYSERR);
+       default:
+               break;
+       }
+       if (waitpid(child, &status, 0) == -1) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "&wait rm");
+       } else if (WIFSIGNALED(status) || WEXITSTATUS(status)) {
+               exitcode = (int)MANDOCLEVEL_SYSERR;
+               say("", "%s: Cannot remove temporary directory", tfn);
+       }
 }
 
 }
 
-static void *
-hash_halloc(size_t sz, void *arg)
+static int
+set_basedir(const char *targetdir, int report_baddir)
 {
 {
+       static char      startdir[PATH_MAX];
+       static int       getcwd_status;  /* 1 = ok, 2 = failure */
+       static int       chdir_status;  /* 1 = changed directory */
+       char            *cp;
 
 
-       return(mandoc_calloc(sz, 1));
-}
-
-static void *
-hash_alloc(size_t sz, void *arg)
-{
+       /*
+        * Remember the original working directory, if possible.
+        * This will be needed if the second or a later directory
+        * on the command line is given as a relative path.
+        * Do not error out if the current directory is not
+        * searchable: Maybe it won't be needed after all.
+        */
+       if (0 == getcwd_status) {
+               if (NULL == getcwd(startdir, sizeof(startdir))) {
+                       getcwd_status = 2;
+                       (void)strlcpy(startdir, strerror(errno),
+                           sizeof(startdir));
+               } else
+                       getcwd_status = 1;
+       }
 
 
-       return(mandoc_malloc(sz));
-}
+       /*
+        * We are leaving the old base directory.
+        * Do not use it any longer, not even for messages.
+        */
+       *basedir = '\0';
 
 
-static void
-hash_free(void *p, size_t sz, void *arg)
-{
+       /*
+        * If and only if the directory was changed earlier and
+        * the next directory to process is given as a relative path,
+        * first go back, or bail out if that is impossible.
+        */
+       if (chdir_status && '/' != *targetdir) {
+               if (2 == getcwd_status) {
+                       exitcode = (int)MANDOCLEVEL_SYSERR;
+                       say("", "getcwd: %s", startdir);
+                       return 0;
+               }
+               if (-1 == chdir(startdir)) {
+                       exitcode = (int)MANDOCLEVEL_SYSERR;
+                       say("", "&chdir %s", startdir);
+                       return 0;
+               }
+       }
 
 
-       free(p);
+       /*
+        * Always resolve basedir to the canonicalized absolute
+        * pathname and append a trailing slash, such that
+        * we can reliably check whether files are inside.
+        */
+       if (NULL == realpath(targetdir, basedir)) {
+               if (report_baddir || errno != ENOENT) {
+                       exitcode = (int)MANDOCLEVEL_BADARG;
+                       say("", "&%s: realpath", targetdir);
+               }
+               return 0;
+       } else if (-1 == chdir(basedir)) {
+               if (report_baddir || errno != ENOENT) {
+                       exitcode = (int)MANDOCLEVEL_BADARG;
+                       say("", "&chdir");
+               }
+               return 0;
+       }
+       chdir_status = 1;
+       cp = strchr(basedir, '\0');
+       if ('/' != cp[-1]) {
+               if (cp - basedir >= PATH_MAX - 1) {
+                       exitcode = (int)MANDOCLEVEL_SYSERR;
+                       say("", "Filename too long");
+                       return 0;
+               }
+               *cp++ = '/';
+               *cp = '\0';
+       }
+       return 1;
 }
 
 }
 
-static int
-path_reset(const char *cwd, int fd, const char *base)
+static void
+say(const char *file, const char *format, ...)
 {
 {
-
-       if (-1 == fchdir(fd)) {
-               perror(cwd);
-               return(0);
-       } else if (-1 == chdir(base)) {
-               perror(base);
-               return(0);
+       va_list          ap;
+       int              use_errno;
+
+       if ('\0' != *basedir)
+               fprintf(stderr, "%s", basedir);
+       if ('\0' != *basedir && '\0' != *file)
+               fputc('/', stderr);
+       if ('\0' != *file)
+               fprintf(stderr, "%s", file);
+
+       use_errno = 1;
+       if (NULL != format) {
+               switch (*format) {
+               case '&':
+                       format++;
+                       break;
+               case '\0':
+                       format = NULL;
+                       break;
+               default:
+                       use_errno = 0;
+                       break;
+               }
        }
        }
-       return(1);
+       if (NULL != format) {
+               if ('\0' != *basedir || '\0' != *file)
+                       fputs(": ", stderr);
+               va_start(ap, format);
+               vfprintf(stderr, format, ap);
+               va_end(ap);
+       }
+       if (use_errno) {
+               if ('\0' != *basedir || '\0' != *file || NULL != format)
+                       fputs(": ", stderr);
+               perror(NULL);
+       } else
+               fputc('\n', stderr);
 }
 }
A git conversion mirror of mandoc.bsd.lv