]> git.cameronkatri.com Git - mandoc.git/blobdiff - read.c
Improve (or rather, rewrite) tbl(7) option parsing.
[mandoc.git] / read.c
diff --git a/read.c b/read.c
index 97d09cdfad37cebaeb5ec500b61832310f6f3798..090f5261e491f403a1c4841091c1971a946b24b7 100644 (file)
--- a/read.c
+++ b/read.c
@@ -1,7 +1,7 @@
-/*     $Id: read.c,v 1.86 2014/09/07 23:25:01 schwarze Exp $ */
+/*     $Id: read.c,v 1.116 2015/01/26 00:57:22 schwarze Exp $ */
 /*
  * Copyright (c) 2008, 2009, 2010, 2011 Kristaps Dzonsons <kristaps@bsd.lv>
- * Copyright (c) 2010-2014 Ingo Schwarze <schwarze@openbsd.org>
+ * Copyright (c) 2010-2015 Ingo Schwarze <schwarze@openbsd.org>
  * Copyright (c) 2010, 2012 Joerg Sonnenberger <joerg@netbsd.org>
  *
  * Permission to use, copy, modify, and distribute this software for any
 #include "libmandoc.h"
 #include "mdoc.h"
 #include "man.h"
-#include "main.h"
 
 #define        REPARSE_LIMIT   1000
 
-struct buf {
-       char             *buf; /* binary input buffer */
-       size_t            sz; /* size of binary buffer */
-};
-
 struct mparse {
        struct man       *pman; /* persistent man parser */
        struct mdoc      *pmdoc; /* persistent mdoc parser */
        struct man       *man; /* man parser */
        struct mdoc      *mdoc; /* mdoc parser */
        struct roff      *roff; /* roff parser (!NULL) */
+       const struct mchars *mchars; /* character table */
        char             *sodest; /* filename pointed to by .so */
        const char       *file; /* filename of current input file */
        struct buf       *primary; /* buffer currently being parsed */
@@ -65,13 +60,15 @@ struct      mparse {
        enum mandoclevel  file_status; /* status of current parse */
        enum mandoclevel  wlevel; /* ignore messages below this */
        int               options; /* parser options */
+       int               filenc; /* encoding of the current file */
        int               reparse_count; /* finite interp. stack */
        int               line; /* line number in the file */
+       pid_t             child; /* the gunzip(1) process */
 };
 
 static void      choose_parser(struct mparse *);
 static void      resize_buf(struct buf *, size_t);
-static void      mparse_buf_r(struct mparse *, struct buf, int);
+static void      mparse_buf_r(struct mparse *, struct buf, size_t, int);
 static int       read_whole_file(struct mparse *, const char *, int,
                                struct buf *, int *);
 static void      mparse_end(struct mparse *);
@@ -83,7 +80,7 @@ static        const enum mandocerr    mandoclimits[MANDOCLEVEL_MAX] = {
        MANDOCERR_WARNING,
        MANDOCERR_WARNING,
        MANDOCERR_ERROR,
-       MANDOCERR_FATAL,
+       MANDOCERR_UNSUPP,
        MANDOCERR_MAX,
        MANDOCERR_MAX
 };
@@ -99,7 +96,6 @@ static        const char * const      mandocerrs[MANDOCERR_MAX] = {
        "lower case character in document title",
        "missing manual section, using \"\"",
        "unknown manual section",
-       "unknown manual volume or arch",
        "missing date, using today's date",
        "cannot parse date, using it verbatim",
        "missing Os macro, using \"\"",
@@ -117,10 +113,13 @@ static    const char * const      mandocerrs[MANDOCERR_MAX] = {
        "sections out of conventional order",
        "duplicate section title",
        "unexpected section",
+       "unusual Xr order",
+       "unusual Xr punctuation",
        "AUTHORS section without An macro",
 
        /* related to macros and nesting */
        "obsolete macro",
+       "macro neither callable nor escaped",
        "skipping paragraph macro",
        "moving paragraph macro out of list",
        "skipping no-space macro",
@@ -146,7 +145,9 @@ static      const char * const      mandocerrs[MANDOCERR_MAX] = {
        "empty list item",
        "missing font type, using \\fR",
        "unknown font type, using \\fR",
+       "nothing follows prefix",
        "missing -std argument, adding it",
+       "missing eqn box, using \"\"",
 
        /* related to bad macro arguments */
        "unterminated quoted argument",
@@ -156,6 +157,8 @@ static      const char * const      mandocerrs[MANDOCERR_MAX] = {
        "skipping duplicate list type",
        "skipping -width argument",
        "unknown AT&T UNIX version",
+       "comma in function argument",
+       "parenthesis in function name",
        "invalid content in Rs block",
        "invalid Boolean argument",
        "unknown font, skipping request",
@@ -175,12 +178,12 @@ static    const char * const      mandocerrs[MANDOCERR_MAX] = {
        "equation scope open on exit",
        "overlapping equation scopes",
        "unexpected end of equation",
-       "equation syntax error",
 
        /* related to tables */
-       "bad table syntax",
-       "bad table option",
-       "bad table layout",
+       "non-alphabetic character in tbl options",
+       "skipping unknown tbl option",
+       "missing tbl option argument",
+       "wrong tbl option argument size",
        "no table layout cells specified",
        "no table data cells specified",
        "ignore data in cell",
@@ -188,44 +191,39 @@ static    const char * const      mandocerrs[MANDOCERR_MAX] = {
        "ignoring extra data cells",
 
        /* related to document structure and macros */
+       NULL,
        "input stack limit exceeded, infinite loop?",
        "skipping bad character",
        "skipping unknown macro",
+       "skipping insecure request",
        "skipping item outside list",
        "skipping column outside column list",
        "skipping end of block that is not open",
+       "fewer RS blocks open, skipping",
        "inserting missing end of block",
        "appending missing end of block",
 
        /* related to request and macro arguments */
        "escaped character not allowed in a name",
        "argument count wrong",
+       "NOT IMPLEMENTED: Bd -file",
        "missing list type, using -item",
        "missing manual name, using \"\"",
        "uname(3) system call failed, using UNKNOWN",
        "unknown standard specifier",
        "skipping request without numeric argument",
+       "NOT IMPLEMENTED: .so with absolute path or \"..\"",
+       ".so request failed",
        "skipping all arguments",
        "skipping excess arguments",
+       "divide by zero",
 
-       "generic fatal error",
-
+       "unsupported feature",
        "input too large",
-       "NOT IMPLEMENTED: Bd -file",
-       "NOT IMPLEMENTED: .so with absolute path or \"..\"",
-       ".so request failed",
-
-       /* system errors */
-       "cannot dup file descriptor",
-       "cannot exec",
-       "gunzip failed with code",
-       "cannot fork",
-       NULL,
-       "cannot open pipe",
-       "cannot read file",
-       "gunzip died from signal",
-       "cannot stat file",
-       "wait failed",
+       "unsupported control character",
+       "unsupported roff request",
+       "unsupported table layout",
+       "ignoring macro in table",
 };
 
 static const char * const      mandoclevels[MANDOCLEVEL_MAX] = {
@@ -233,7 +231,7 @@ static      const char * const      mandoclevels[MANDOCLEVEL_MAX] = {
        "RESERVED",
        "WARNING",
        "ERROR",
-       "FATAL",
+       "UNSUPP",
        "BADARG",
        "SYSERR"
 };
@@ -294,43 +292,55 @@ choose_parser(struct mparse *curp)
        /* Fall back to man(7) as a last resort. */
 
        if (NULL == curp->pman)
-               curp->pman = man_alloc(curp->roff, curp,
+               curp->pman = man_alloc(
+                   curp->roff, curp, curp->defos,
                    MPARSE_QUICK & curp->options ? 1 : 0);
        assert(curp->pman);
        curp->man = curp->pman;
 }
 
 /*
- * Main parse routine for an opened file.  This is called for each
- * opened file and simply loops around the full input file, possibly
- * nesting (i.e., with `so').
+ * Main parse routine for a buffer.
+ * It assumes encoding and line numbering are already set up.
+ * It can recurse directly (for invocations of user-defined
+ * macros, inline equations, and input line traps)
+ * and indirectly (for .so file inclusion).
  */
 static void
-mparse_buf_r(struct mparse *curp, struct buf blk, int start)
+mparse_buf_r(struct mparse *curp, struct buf blk, size_t i, int start)
 {
        const struct tbl_span   *span;
        struct buf       ln;
+       const char      *save_file;
+       char            *cp;
+       size_t           pos; /* byte number in the ln buffer */
        enum rofferr     rr;
-       int              i, of, rc;
-       int              pos; /* byte number in the ln buffer */
+       int              of;
        int              lnn; /* line number in the real file */
+       int              fd;
+       pid_t            save_child;
        unsigned char    c;
 
-       memset(&ln, 0, sizeof(struct buf));
+       memset(&ln, 0, sizeof(ln));
 
        lnn = curp->line;
        pos = 0;
 
-       for (i = 0; i < (int)blk.sz; ) {
+       while (i < blk.sz) {
                if (0 == pos && '\0' == blk.buf[i])
                        break;
 
                if (start) {
                        curp->line = lnn;
                        curp->reparse_count = 0;
+
+                       if (lnn < 3 &&
+                           curp->filenc & MPARSE_UTF8 &&
+                           curp->filenc & MPARSE_LATIN1)
+                               curp->filenc = preconv_cue(&blk, i);
                }
 
-               while (i < (int)blk.sz && (start || '\0' != blk.buf[i])) {
+               while (i < blk.sz && (start || blk.buf[i] != '\0')) {
 
                        /*
                         * When finding an unescaped newline character,
@@ -338,7 +348,7 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
                         * Skip a preceding carriage return, if any.
                         */
 
-                       if ('\r' == blk.buf[i] && i + 1 < (int)blk.sz &&
+                       if ('\r' == blk.buf[i] && i + 1 < blk.sz &&
                            '\n' == blk.buf[i + 1])
                                ++i;
                        if ('\n' == blk.buf[i]) {
@@ -348,29 +358,38 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
                        }
 
                        /*
-                        * Make sure we have space for at least
-                        * one backslash and one other character
-                        * and the trailing NUL byte.
+                        * Make sure we have space for the worst
+                        * case of 11 bytes: "\\[u10ffff]\0"
                         */
 
-                       if (pos + 2 >= (int)ln.sz)
+                       if (pos + 11 > ln.sz)
                                resize_buf(&ln, 256);
 
                        /*
-                        * Warn about bogus characters.  If you're using
-                        * non-ASCII encoding, you're screwing your
-                        * readers.  Since I'd rather this not happen,
-                        * I'll be helpful and replace these characters
-                        * with "?", so we don't display gibberish.
-                        * Note to manual writers: use special characters.
+                        * Encode 8-bit input.
                         */
 
-                       c = (unsigned char) blk.buf[i];
+                       c = blk.buf[i];
+                       if (c & 0x80) {
+                               if ( ! (curp->filenc && preconv_encode(
+                                   &blk, &i, &ln, &pos, &curp->filenc))) {
+                                       mandoc_vmsg(MANDOCERR_CHAR_BAD, curp,
+                                           curp->line, pos, "0x%x", c);
+                                       ln.buf[pos++] = '?';
+                                       i++;
+                               }
+                               continue;
+                       }
 
-                       if ( ! (isascii(c) &&
-                           (isgraph(c) || isblank(c)))) {
-                               mandoc_vmsg(MANDOCERR_BADCHAR, curp,
-                                   curp->line, pos, "0x%x", c);
+                       /*
+                        * Exclude control characters.
+                        */
+
+                       if (c == 0x7f || (c < 0x20 && c != 0x09)) {
+                               mandoc_vmsg(c == 0x00 || c == 0x04 ||
+                                   c > 0x0a ? MANDOCERR_CHAR_BAD :
+                                   MANDOCERR_CHAR_UNSUPP,
+                                   curp, curp->line, pos, "0x%x", c);
                                i++;
                                ln.buf[pos++] = '?';
                                continue;
@@ -378,7 +397,7 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
 
                        /* Trailing backslash = a plain char. */
 
-                       if ('\\' != blk.buf[i] || i + 1 == (int)blk.sz) {
+                       if (blk.buf[i] != '\\' || i + 1 == blk.sz) {
                                ln.buf[pos++] = blk.buf[i++];
                                continue;
                        }
@@ -390,7 +409,7 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
                         * skip that one as well.
                         */
 
-                       if ('\r' == blk.buf[i + 1] && i + 2 < (int)blk.sz &&
+                       if ('\r' == blk.buf[i + 1] && i + 2 < blk.sz &&
                            '\n' == blk.buf[i + 2])
                                ++i;
                        if ('\n' == blk.buf[i + 1]) {
@@ -402,7 +421,7 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
                        if ('"' == blk.buf[i + 1] || '#' == blk.buf[i + 1]) {
                                i += 2;
                                /* Comment, skip to end of line */
-                               for (; i < (int)blk.sz; ++i) {
+                               for (; i < blk.sz; ++i) {
                                        if ('\n' == blk.buf[i]) {
                                                ++i;
                                                ++lnn;
@@ -426,7 +445,7 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
 
                        if ( ! (isascii(c) &&
                            (isgraph(c) || isblank(c)))) {
-                               mandoc_vmsg(MANDOCERR_BADCHAR, curp,
+                               mandoc_vmsg(MANDOCERR_CHAR_BAD, curp,
                                    curp->line, pos, "0x%x", c);
                                i += 2;
                                ln.buf[pos++] = '?';
@@ -439,7 +458,7 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
                        ln.buf[pos++] = blk.buf[i++];
                }
 
-               if (pos >= (int)ln.sz)
+               if (pos >= ln.sz)
                        resize_buf(&ln, 256);
 
                ln.buf[pos] = '\0';
@@ -476,32 +495,28 @@ mparse_buf_r(struct mparse *curp, struct buf blk, int start)
                                [curp->secondary->sz] = '\0';
                }
 rerun:
-               rr = roff_parseln(curp->roff, curp->line,
-                   &ln.buf, &ln.sz, of, &of);
+               rr = roff_parseln(curp->roff, curp->line, &ln, &of);
 
                switch (rr) {
                case ROFF_REPARSE:
                        if (REPARSE_LIMIT >= ++curp->reparse_count)
-                               mparse_buf_r(curp, ln, 0);
+                               mparse_buf_r(curp, ln, of, 0);
                        else
                                mandoc_msg(MANDOCERR_ROFFLOOP, curp,
                                    curp->line, pos, NULL);
                        pos = 0;
                        continue;
                case ROFF_APPEND:
-                       pos = (int)strlen(ln.buf);
+                       pos = strlen(ln.buf);
                        continue;
                case ROFF_RERUN:
                        goto rerun;
                case ROFF_IGN:
                        pos = 0;
                        continue;
-               case ROFF_ERR:
-                       assert(MANDOCLEVEL_FATAL <= curp->file_status);
-                       break;
                case ROFF_SO:
-                       if (0 == (MPARSE_SO & curp->options) &&
-                           (i >= (int)blk.sz || '\0' == blk.buf[i])) {
+                       if ( ! (curp->options & MPARSE_SO) &&
+                           (i >= blk.sz || blk.buf[i] == '\0')) {
                                curp->sodest = mandoc_strdup(ln.buf + of);
                                free(ln.buf);
                                return;
@@ -513,27 +528,32 @@ rerun:
                         */
                        if (curp->secondary)
                                curp->secondary->sz -= pos + 1;
-                       mparse_readfd(curp, -1, ln.buf + of);
-                       if (MANDOCLEVEL_FATAL <= curp->file_status) {
+                       save_file = curp->file;
+                       save_child = curp->child;
+                       if (mparse_open(curp, &fd, ln.buf + of) ==
+                           MANDOCLEVEL_OK) {
+                               mparse_readfd(curp, fd, ln.buf + of);
+                               curp->file = save_file;
+                       } else {
+                               curp->file = save_file;
                                mandoc_vmsg(MANDOCERR_SO_FAIL,
                                    curp, curp->line, pos,
                                    ".so %s", ln.buf + of);
-                               break;
+                               ln.sz = mandoc_asprintf(&cp,
+                                   ".sp\nSee the file %s.\n.sp",
+                                   ln.buf + of);
+                               free(ln.buf);
+                               ln.buf = cp;
+                               of = 0;
+                               mparse_buf_r(curp, ln, of, 0);
                        }
+                       curp->child = save_child;
                        pos = 0;
                        continue;
                default:
                        break;
                }
 
-               /*
-                * If we encounter errors in the recursive parse, make
-                * sure we don't continue parsing.
-                */
-
-               if (MANDOCLEVEL_FATAL <= curp->file_status)
-                       break;
-
                /*
                 * If input parsers have not been allocated, do so now.
                 * We keep these instanced between parsers, but set them
@@ -553,34 +573,21 @@ rerun:
                 * Do the same for ROFF_EQN.
                 */
 
-               rc = -1;
-
-               if (ROFF_TBL == rr)
-                       while (NULL != (span = roff_span(curp->roff))) {
-                               rc = curp->man ?
-                                   man_addspan(curp->man, span) :
-                                   mdoc_addspan(curp->mdoc, span);
-                               if (0 == rc)
-                                       break;
-                       }
-               else if (ROFF_EQN == rr)
-                       rc = curp->mdoc ?
-                           mdoc_addeqn(curp->mdoc,
-                               roff_eqn(curp->roff)) :
-                           man_addeqn(curp->man,
-                               roff_eqn(curp->roff));
-               else if (curp->man || curp->mdoc)
-                       rc = curp->man ?
-                           man_parseln(curp->man,
-                               curp->line, ln.buf, of) :
-                           mdoc_parseln(curp->mdoc,
-                               curp->line, ln.buf, of);
-
-               if (0 == rc) {
-                       assert(MANDOCLEVEL_FATAL <= curp->file_status);
-                       break;
-               } else if (2 == rc)
-                       break;
+               if (rr == ROFF_TBL) {
+                       while ((span = roff_span(curp->roff)) != NULL)
+                               if (curp->man == NULL)
+                                       mdoc_addspan(curp->mdoc, span);
+                               else
+                                       man_addspan(curp->man, span);
+               } else if (rr == ROFF_EQN) {
+                       if (curp->man == NULL)
+                               mdoc_addeqn(curp->mdoc, roff_eqn(curp->roff));
+                       else
+                               man_addeqn(curp->man, roff_eqn(curp->roff));
+               } else if ((curp->man == NULL ?
+                   mdoc_parseln(curp->mdoc, curp->line, ln.buf, of) :
+                   man_parseln(curp->man, curp->line, ln.buf, of)) == 2)
+                               break;
 
                /* Temporary buffers typically are not full. */
 
@@ -605,11 +612,8 @@ read_whole_file(struct mparse *curp, const char *file, int fd,
 #if HAVE_MMAP
        struct stat      st;
        if (-1 == fstat(fd, &st)) {
-               curp->file_status = MANDOCLEVEL_SYSERR;
-               if (curp->mmsg)
-                       (*curp->mmsg)(MANDOCERR_SYSSTAT, curp->file_status,
-                           file, 0, 0, strerror(errno));
-               return(0);
+               perror(file);
+               exit((int)MANDOCLEVEL_SYSERR);
        }
 
        /*
@@ -621,10 +625,7 @@ read_whole_file(struct mparse *curp, const char *file, int fd,
 
        if (S_ISREG(st.st_mode)) {
                if (st.st_size >= (1U << 31)) {
-                       curp->file_status = MANDOCLEVEL_FATAL;
-                       if (curp->mmsg)
-                               (*curp->mmsg)(MANDOCERR_TOOLARGE,
-                                   curp->file_status, file, 0, 0, NULL);
+                       mandoc_msg(MANDOCERR_TOOLARGE, curp, 0, 0, NULL);
                        return(0);
                }
                *with_mmap = 1;
@@ -647,11 +648,8 @@ read_whole_file(struct mparse *curp, const char *file, int fd,
        for (;;) {
                if (off == fb->sz) {
                        if (fb->sz == (1U << 31)) {
-                               curp->file_status = MANDOCLEVEL_FATAL;
-                               if (curp->mmsg)
-                                       (*curp->mmsg)(MANDOCERR_TOOLARGE,
-                                           curp->file_status,
-                                           file, 0, 0, NULL);
+                               mandoc_msg(MANDOCERR_TOOLARGE, curp,
+                                   0, 0, NULL);
                                break;
                        }
                        resize_buf(fb, 65536);
@@ -662,12 +660,8 @@ read_whole_file(struct mparse *curp, const char *file, int fd,
                        return(1);
                }
                if (ssz == -1) {
-                       curp->file_status = MANDOCLEVEL_SYSERR;
-                       if (curp->mmsg)
-                               (*curp->mmsg)(MANDOCERR_SYSREAD,
-                                   curp->file_status, file, 0, 0,
-                                   strerror(errno));
-                       break;
+                       perror(file);
+                       exit((int)MANDOCLEVEL_SYSERR);
                }
                off += (size_t)ssz;
        }
@@ -681,9 +675,6 @@ static void
 mparse_end(struct mparse *curp)
 {
 
-       if (MANDOCLEVEL_FATAL <= curp->file_status)
-               return;
-
        if (curp->mdoc == NULL &&
            curp->man == NULL &&
            curp->sodest == NULL) {
@@ -691,22 +682,16 @@ mparse_end(struct mparse *curp)
                        curp->mdoc = curp->pmdoc;
                else {
                        if (curp->pman == NULL)
-                               curp->pman = man_alloc(curp->roff, curp,
+                               curp->pman = man_alloc(
+                                   curp->roff, curp, curp->defos,
                                    curp->options & MPARSE_QUICK ? 1 : 0);
                        curp->man = curp->pman;
                }
        }
-
-       if (curp->mdoc && ! mdoc_endparse(curp->mdoc)) {
-               assert(MANDOCLEVEL_FATAL <= curp->file_status);
-               return;
-       }
-
-       if (curp->man && ! man_endparse(curp->man)) {
-               assert(MANDOCLEVEL_FATAL <= curp->file_status);
-               return;
-       }
-
+       if (curp->mdoc)
+               mdoc_endparse(curp->mdoc);
+       if (curp->man)
+               man_endparse(curp->man);
        roff_endparse(curp->roff);
 }
 
@@ -715,6 +700,7 @@ mparse_parse_buffer(struct mparse *curp, struct buf blk, const char *file)
 {
        struct buf      *svprimary;
        const char      *svfile;
+       size_t           offset;
        static int       recursion_depth;
 
        if (64 < recursion_depth) {
@@ -730,9 +716,19 @@ mparse_parse_buffer(struct mparse *curp, struct buf blk, const char *file)
        curp->line = 1;
        recursion_depth++;
 
-       mparse_buf_r(curp, blk, 1);
+       /* Skip an UTF-8 byte order mark. */
+       if (curp->filenc & MPARSE_UTF8 && blk.sz > 2 &&
+           (unsigned char)blk.buf[0] == 0xef &&
+           (unsigned char)blk.buf[1] == 0xbb &&
+           (unsigned char)blk.buf[2] == 0xbf) {
+               offset = 3;
+               curp->filenc &= ~MPARSE_LATIN1;
+       } else
+               offset = 0;
+
+       mparse_buf_r(curp, blk, offset, 1);
 
-       if (0 == --recursion_depth && MANDOCLEVEL_FATAL > curp->file_status)
+       if (--recursion_depth == 0)
                mparse_end(curp);
 
        curp->primary = svprimary;
@@ -740,151 +736,145 @@ mparse_parse_buffer(struct mparse *curp, struct buf blk, const char *file)
 }
 
 enum mandoclevel
-mparse_readmem(struct mparse *curp, const void *buf, size_t len,
+mparse_readmem(struct mparse *curp, void *buf, size_t len,
                const char *file)
 {
        struct buf blk;
 
-       blk.buf = UNCONST(buf);
+       blk.buf = buf;
        blk.sz = len;
 
        mparse_parse_buffer(curp, blk, file);
        return(curp->file_status);
 }
 
+/*
+ * Read the whole file into memory and call the parsers.
+ * Called recursively when an .so request is encountered.
+ */
 enum mandoclevel
 mparse_readfd(struct mparse *curp, int fd, const char *file)
 {
        struct buf       blk;
        int              with_mmap;
-
-       if (-1 == fd && -1 == (fd = open(file, O_RDONLY, 0))) {
-               curp->file_status = MANDOCLEVEL_SYSERR;
-               if (curp->mmsg)
-                       (*curp->mmsg)(MANDOCERR_SYSOPEN,
-                           curp->file_status,
-                           file, 0, 0, strerror(errno));
-               goto out;
-       }
-
-       /*
-        * Run for each opened file; may be called more than once for
-        * each full parse sequence if the opened file is nested (i.e.,
-        * from `so').  Simply sucks in the whole file and moves into
-        * the parse phase for the file.
-        */
-
-       if ( ! read_whole_file(curp, file, fd, &blk, &with_mmap))
-               goto out;
-
-       mparse_parse_buffer(curp, blk, file);
-
+       int              save_filenc;
+
+       if (read_whole_file(curp, file, fd, &blk, &with_mmap)) {
+               save_filenc = curp->filenc;
+               curp->filenc = curp->options &
+                   (MPARSE_UTF8 | MPARSE_LATIN1);
+               mparse_parse_buffer(curp, blk, file);
+               curp->filenc = save_filenc;
 #if HAVE_MMAP
-       if (with_mmap)
-               munmap(blk.buf, blk.sz);
-       else
+               if (with_mmap)
+                       munmap(blk.buf, blk.sz);
+               else
 #endif
-               free(blk.buf);
+                       free(blk.buf);
+       }
 
-       if (STDIN_FILENO != fd && -1 == close(fd))
+       if (fd != STDIN_FILENO && close(fd) == -1)
                perror(file);
-out:
+
+       mparse_wait(curp);
        return(curp->file_status);
 }
 
 enum mandoclevel
-mparse_open(struct mparse *curp, int *fd, const char *file,
-       pid_t *child_pid)
+mparse_open(struct mparse *curp, int *fd, const char *file)
 {
        int               pfd[2];
+       int               save_errno;
        char             *cp;
-       enum mandocerr    err;
 
-       pfd[1] = -1;
        curp->file = file;
+
+       /* Unless zipped, try to just open the file. */
+
        if ((cp = strrchr(file, '.')) == NULL ||
            strcmp(cp + 1, "gz")) {
-               *child_pid = 0;
-               if ((*fd = open(file, O_RDONLY)) == -1) {
-                       err = MANDOCERR_SYSOPEN;
-                       goto out;
-               }
-               return(MANDOCLEVEL_OK);
+               curp->child = 0;
+               if ((*fd = open(file, O_RDONLY)) != -1)
+                       return(MANDOCLEVEL_OK);
+
+               /* Open failed; try to append ".gz". */
+
+               mandoc_asprintf(&cp, "%s.gz", file);
+               file = cp;
+       } else
+               cp = NULL;
+
+       /* Before forking, make sure the file can be read. */
+
+       save_errno = errno;
+       if (access(file, R_OK) == -1) {
+               if (cp != NULL)
+                       errno = save_errno;
+               free(cp);
+               *fd = -1;
+               curp->child = 0;
+               mandoc_msg(MANDOCERR_FILE, curp, 0, 0, strerror(errno));
+               return(MANDOCLEVEL_ERROR);
        }
 
+       /* Run gunzip(1). */
+
        if (pipe(pfd) == -1) {
-               err = MANDOCERR_SYSPIPE;
-               goto out;
+               perror("pipe");
+               exit((int)MANDOCLEVEL_SYSERR);
        }
 
-       switch (*child_pid = fork()) {
+       switch (curp->child = fork()) {
        case -1:
-               err = MANDOCERR_SYSFORK;
-               close(pfd[0]);
-               close(pfd[1]);
-               pfd[1] = -1;
-               break;
+               perror("fork");
+               exit((int)MANDOCLEVEL_SYSERR);
        case 0:
                close(pfd[0]);
                if (dup2(pfd[1], STDOUT_FILENO) == -1) {
-                       err = MANDOCERR_SYSDUP;
-                       break;
+                       perror("dup");
+                       exit((int)MANDOCLEVEL_SYSERR);
                }
                execlp("gunzip", "gunzip", "-c", file, NULL);
-               err = MANDOCERR_SYSEXEC;
-               break;
+               perror("exec");
+               exit((int)MANDOCLEVEL_SYSERR);
        default:
                close(pfd[1]);
                *fd = pfd[0];
                return(MANDOCLEVEL_OK);
        }
-
-out:
-       *fd = -1;
-       *child_pid = 0;
-       curp->file_status = MANDOCLEVEL_SYSERR;
-       if (curp->mmsg)
-               (*curp->mmsg)(err, curp->file_status, file,
-                   0, 0, strerror(errno));
-       if (pfd[1] != -1)
-               exit(1);
-       return(curp->file_status);
 }
 
 enum mandoclevel
-mparse_wait(struct mparse *curp, pid_t child_pid)
+mparse_wait(struct mparse *curp)
 {
        int       status;
 
-       if (waitpid(child_pid, &status, 0) == -1) {
-               mandoc_msg(MANDOCERR_SYSWAIT, curp, 0, 0,
-                   strerror(errno));
-               curp->file_status = MANDOCLEVEL_SYSERR;
-               return(curp->file_status);
+       if (curp->child == 0)
+               return(MANDOCLEVEL_OK);
+
+       if (waitpid(curp->child, &status, 0) == -1) {
+               perror("wait");
+               exit((int)MANDOCLEVEL_SYSERR);
        }
        if (WIFSIGNALED(status)) {
-               mandoc_vmsg(MANDOCERR_SYSSIG, curp, 0, 0,
-                   "%d", WTERMSIG(status));
-               curp->file_status = MANDOCLEVEL_SYSERR;
-               return(curp->file_status);
+               mandoc_vmsg(MANDOCERR_FILE, curp, 0, 0,
+                   "gunzip died from signal %d", WTERMSIG(status));
+               return(MANDOCLEVEL_ERROR);
        }
        if (WEXITSTATUS(status)) {
-               mandoc_vmsg(MANDOCERR_SYSEXIT, curp, 0, 0,
-                   "%d", WEXITSTATUS(status));
-               curp->file_status = MANDOCLEVEL_SYSERR;
-               return(curp->file_status);
+               mandoc_vmsg(MANDOCERR_FILE, curp, 0, 0,
+                   "gunzip failed with code %d", WEXITSTATUS(status));
+               return(MANDOCLEVEL_ERROR);
        }
        return(MANDOCLEVEL_OK);
 }
 
 struct mparse *
-mparse_alloc(int options, enum mandoclevel wlevel,
-               mandocmsg mmsg, const char *defos)
+mparse_alloc(int options, enum mandoclevel wlevel, mandocmsg mmsg,
+    const struct mchars *mchars, const char *defos)
 {
        struct mparse   *curp;
 
-       assert(wlevel <= MANDOCLEVEL_FATAL);
-
        curp = mandoc_calloc(1, sizeof(struct mparse));
 
        curp->options = options;
@@ -892,13 +882,15 @@ mparse_alloc(int options, enum mandoclevel wlevel,
        curp->mmsg = mmsg;
        curp->defos = defos;
 
-       curp->roff = roff_alloc(curp, options);
+       curp->mchars = mchars;
+       curp->roff = roff_alloc(curp, curp->mchars, options);
        if (curp->options & MPARSE_MDOC)
                curp->pmdoc = mdoc_alloc(
                    curp->roff, curp, curp->defos,
                    curp->options & MPARSE_QUICK ? 1 : 0);
        if (curp->options & MPARSE_MAN)
-               curp->pman = man_alloc(curp->roff, curp,
+               curp->pman = man_alloc(
+                   curp->roff, curp, curp->defos,
                    curp->options & MPARSE_QUICK ? 1 : 0);
 
        return(curp);
@@ -979,11 +971,11 @@ mandoc_msg(enum mandocerr er, struct mparse *m,
 {
        enum mandoclevel level;
 
-       level = MANDOCLEVEL_FATAL;
+       level = MANDOCLEVEL_UNSUPP;
        while (er < mandoclimits[level])
                level--;
 
-       if (level < m->wlevel)
+       if (level < m->wlevel && er != MANDOCERR_FILE)
                return;
 
        if (m->mmsg)