Ingo Schwarze [Sat, 20 Aug 2016 14:43:50 +0000 (14:43 +0000)]
If a column list starts with implicit rows (that is, rows without .It)
and roff-level nodes (e.g. tbl or eqn) follow, don't run into an
assertion. Instead, wrap the roff-level nodes in their own row.
Issue found by tb@ with afl(1).
Ingo Schwarze [Wed, 17 Aug 2016 20:46:56 +0000 (20:46 +0000)]
When the content of a manual page does not specify a section, the
empty string got added to the list of sections, breaking the database
format slightly and causing the page to not be considered part of
any section, not even if a section could be deduced from the directory
or from the file name.
Bug found due to the bogus pcredemo(3) "manual" in the pcre-8.38p0 package.
Ingo Schwarze [Wed, 17 Aug 2016 18:59:37 +0000 (18:59 +0000)]
When reading back a mandoc.db(5) file in order to apply incremental
changes, do not prepend a stray NAME_FILE (0x10) byte to the first
names of pages.
Bug found while investigating another issue reported by sthen@.
Ingo Schwarze [Wed, 17 Aug 2016 18:10:39 +0000 (18:10 +0000)]
Make sure manuals in architecture-independent directories are treated
as architecture-independent even if they abuse the third (architecture)
argument of the .Dt macro for random stuff like "freetds reference manual".
While the .Dt syntax is not the same as the .TH syntax in man(7),
punishing offenders by treating them as architecture-dependent and
hence completely excluding them from searches is too severe.
Problem reported by sthen@.
Ingo Schwarze [Thu, 11 Aug 2016 13:30:25 +0000 (13:30 +0000)]
Even after switching from a pending head to the body, we have to
continue scanning upwards, because the enclosing block might already
be pending as well, e.g. .Bl .Bl .It Bo .El .It.
Tree corruption leading to a later NULL deref found by tb@ with afl(1).
Ingo Schwarze [Thu, 11 Aug 2016 10:47:16 +0000 (10:47 +0000)]
If a .Bd display is on the one hand doomed to be deleted because
it has no type, but is on the other hand breaking another block,
delete its end marker as well, or the end marker may remain behind
as an orphan, triggering an assertion in the terminal formatter.
Problem found by tb@ with afl(1).
Ingo Schwarze [Wed, 10 Aug 2016 20:17:50 +0000 (20:17 +0000)]
Don't deref NULL if the only child of the first .Sh is an empty
in-line macro, and don't printf("%s", NULL) if the first child
of the first .Sh is a macro; again found by tb@ with afl(1).
(No, you should never use macros in any .Sh at all, please.)
Ingo Schwarze [Wed, 10 Aug 2016 12:50:24 +0000 (12:50 +0000)]
When trying to figure out which C compiler make(1) wants to use,
pass it the POSIX -s option. On most systems, this won't make a
difference, but Bdale Garbee reported that the make(1) on his Debian
system, most likely some version of gmake, breaks Makefile.local
by printing some 'entering directory' messages. I failed to reproduce
and Bdale didn't report back, but judging from gmake source code,
this is likely to help and unlikely to do harm elsewhere.
Ingo Schwarze [Wed, 10 Aug 2016 12:06:41 +0000 (12:06 +0000)]
When validating a .Bl list that defaults to -item for want of a type,
don't let a subsequent -width access mdoc_argnames[] out of bounds.
Found by tb@ with afl(1).
Ingo Schwarze [Wed, 10 Aug 2016 11:03:43 +0000 (11:03 +0000)]
Fix assertion failures caused by whitespace inside \o'' (overstrike)
sequences that jsg@ found with afl(1):
* Avoid writing \t\b in term.c.
* Handle trailing \b in term_ps.c.
Ingo Schwarze [Fri, 5 Aug 2016 23:15:08 +0000 (23:15 +0000)]
The concept of endianness seems to be somewhat newfangled, so the
respective conversion functions are not yet properly standardized.
Rumour has it that POSIX is working on it, though.
For now, sprinkle some configuration glue.
Ingo Schwarze [Thu, 4 Aug 2016 09:33:57 +0000 (09:33 +0000)]
Fix an assertion failure that happened when trying to add a page
with makewhatis -d to a completely empty database.
Reported by Mark Patruck <mark at wrapped dot cx>, thanks!
Ingo Schwarze [Tue, 2 Aug 2016 11:09:46 +0000 (11:09 +0000)]
POSIX allows PATH_MAX to not be defined, meaning "unlimited".
Found by Aaron M. Ucko <amu at alum dot mit dot edu> on the GNU Hurd,
via Bdale Garbee, https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=829624
Also add EFTYPE at two places where it was forgotten.
Some base system pages, for example perl(1), contain non-ASCII
characters in their source code, so switch on charset autodetection
in the same way as in man(1) itself.
Issue reported by Pavan Maddamsetti at gmail dot com on bugs@.
Autodetect a suitable locale for -Tutf8 mode,
and allow overriding it manually.
Based on a patch from Svyatoslav Mishyn <juef at openmailbox dot org>
tweaked by me.
The idea originally came up in a conversation with Markus Waldeck.
No need to populate the TYPE_arch and TYPE_sec bits, the information
is provided directly to dba_page_add() in dbadd_mlink()
and to dba_page_new() in dbadd().
No need for a dedicated loop for NAME_FILE.
It's done in dbadd_mlink() anyway.
In this context, also record section numbers taken from filenames
and from .Dt and .TH macros, architectures taken from .Dt macros,
and fix the filtering of duplicate filename entries.
Now that our man.conf(5) format is mature and extremely simple,
delete manpath(1) support. With the mandoc-based man(1), manpath(1)
is utterly useless. Just set MANPATH_DEFAULT in configure.local
for sane operating system defaults, use man.conf(5) for machine-
specific modifications, and use ${MANPATH}, -m, and -M for user
preferences.
Remove the dependency on SQLite without loss of functionality.
Stop supporting systems that don't have mmap(3).
Drop the obsolete names_check() now that we deleted MLINKS.
Since the mdoc/man parser unification, the parser is always allocated
in mparse_alloc(), so delete all the curp->man == NULL checks.
Triggered by a patch from Christos Zoulas suggesting to add
yet another such check.
To remove the const qualifier from a pointer to an object - either
because we know it is actually mutable or because we are passing
it to a function that doesn't accept a const object but won't
actually attempt to modify it - simply casting from (const type *)
to (type *) is legal C and clearly expresses the intent.
So get rid of the obfuscating UNCONST macro.
Basic idea discussed with guenther@.
Add support for Mac OS X's sandbox_init(3) sandbox functionality, which
is marked as DEPRECATED in OS X after 2011 or so, but has not been
removed and has no replacement.
Make all components of the URI individually optional,
independent of each other, as in:
http://man.openbsd.org[/manpath][/mansec][/arch]/name[.sec]
The restrictions in the past kept confusing people.
Triggered by a question from RafaelNeves at gmail dot com.
On OpenBSD, more(1) is actually less(1) with a small number of
compatibility features - so we can safely use more -T on OpenBSD.
But don't do that in the portable version: more -T is unlikely
to work elsewhere.
Issue reported by Svyatoslav Mishyn <juef at openmailbox dot org>.
Fix a nasty typo that prevented .so links to gziped manuals
from working in the absence of a mandoc.db(5) database.
Found the hard way by Svyatoslav Mishyn on Crux Linux.
Simplify the code and the server setup by deleting the pseudo-manpath
"mandoc" that was used for man.cgi(8) documentation and by assuming
that the apropos(1) and man.cgi(8) manuals are simply installed in
the default manpath. Even though man.cgi(8) is not installed by
default when installing OpenBSD, it is easy to copy it into the
default manpath used for man.cgi(8).
Idea found when considering a question asked by wrant dot com.
Do not treat PATH_INFO as a complete path if it doesn't contain
a manpath. For example, this makes http://man.openbsd.org/mandoc
work as expected.
Bug reported by tb@, reminded by Svyatoslav Mishyn.
ISO C99 7.19.2.5 doesn't like mixing putchar(3) and putwchar(3) on
the same stream, and actually, it fails spectacularly on glibc.
Portability issue pointed out by Svyatoslav Mishyn <juef at openmailbox
dot org> after testing on Void Linux.
POSIX requires that a process calling tcsetpgrp(3) from the background
gets a SIGTTOU signal. In that case, do not stop.
Portability issue found while testing on commercial Solaris 9/10/11.
Thanks to opencsw.org for providing me with a testing environment.
Update and simplify the documentation of the -s option,
which was forgotten when implementing the new man.conf(5) format.
The outdated information was originally pointed out
by Andy Bradford <amb dash openbsd at bradfords dot org> on misc@.
OK jmc@
Ingo Schwarze [Sat, 28 May 2016 13:40:48 +0000 (13:40 +0000)]
Simplify search form: minus two visible control elements, minus
one table, minus twenty lines of code, no loss of functionality.
No idea why i didn't do this earlier...
Ingo Schwarze [Wed, 18 May 2016 23:51:16 +0000 (23:51 +0000)]
Delete useless variables that could sneak into the Makefile
behind the user's back, dangerously bypassing ./configure.
Leakage reported by Peter Bray <pdb_ml at yahoo dot com dot au>.
Ingo Schwarze [Wed, 18 May 2016 21:37:04 +0000 (21:37 +0000)]
Various people (among others Maxim Belooussov and Carsten Kunze)
reported that the build system still assumed that ohash is only
needed if sqlite3 is also in use, which is no longer true:
The ohash library is now required no matter what.
Rework sqlite3 and ohash library autodetection
such that both work independently of each other.
Provide LDADD for additional linker flags.
Add some missing variables to configure.local.example.
Only focus on the query input box when no manual page is displayed,
that is, for the index page, for the noresult page, and for the
result of an apropos(1) query with more than one page.
As noted by bentley@, when a manual page is displayed, it is more
important that people can quickly use the space bar for paging and
Ctrl-F for searching.
Rename five static functions to make the classification of functions
as parsers, page generators, and result generators more obvious.
No functional change.
If PATH_INFO contains a complete and correct path to a manual page
file, for example "/OpenBSD-5.9/man2/pledge.2", no database query
is needed and the file is delivered directly.
But even in this case, let's parse the PATH_INFO and fill the query
structure such that the search form at the top of the result page
gets pre-filled with useful values.
It could occasionally happen that the child process spawned less(1)
before the parent process passed the control of the terminal to the
child, and in that case, less(1) sometimes complained "Stopped (tty
output)". Issue reported by naddy@.
Give manuals in purely numerical sections priority over manuals of
the same name in sections with an alphabetical suffix; same logic
as in main.c rev. 1.264.
Give manuals in purely numerical sections priority over manuals of
the same name in sections with an alphabetical suffix (on OpenBSD,
mostly 3p), restoring behaviour of the traditional BSD man(1) that
got lost in the switch to the mandoc-based implementation.
Issue reported by jsg@, using an idea by mikeb@ for the solution,
and at least afresh1@ and jasper@ also seem in favour of the direction.
Ingo Schwarze [Fri, 8 Jan 2016 17:48:09 +0000 (17:48 +0000)]
Delete the redundant "nchild" member of struct roff_node, replacing
most uses by one, a few by two pointer checks, and only one by a
tiny loop - not only making data smaller, but code shorter as well.
This gets rid of an implicit invariant that confused both static
analysis tools and human auditors. No functional change.
Ingo Schwarze [Fri, 8 Jan 2016 15:02:54 +0000 (15:02 +0000)]
Prefer warn(3) over perror(3) at the few places where it was used.
It is useful to see the program name, and we have err.h compat in place anyway.
Suggested by Christos Zoulas (NetBSD).
Ingo Schwarze [Fri, 8 Jan 2016 02:53:13 +0000 (02:53 +0000)]
Simplify the mparse_open() interface.
Just return the file descriptor or -1 on error;
there is just one kind of error anyway.
Suggested by Christos Zoulas (NetBSD).
Ingo Schwarze [Fri, 8 Jan 2016 02:13:39 +0000 (02:13 +0000)]
It was very surprising that a function called mparse_readfd()
closed the file descriptor passed to it after completing its work,
in particular considering the fact that it required its callers
to call open(2) or mparse_open() beforehand.
Change mparse_readfd() to not call close(2) and change the callers
to call close(2) afterwards, more or less bringing open and close
to the same level of the code and making review easier. Note that
man.cgi(8) already did that, even though it was wrong in the past.
Small restructuring suggested by Christos Zoulas (NetBSD).
Ingo Schwarze [Fri, 8 Jan 2016 00:50:45 +0000 (00:50 +0000)]
The root of an .EQ tree is always EQN_ROOT, never EQN_LIST,
so delete a redundant NULL check that confused Coverity in CID 1257471;
issue reported by wiz@, patch differs from what christos@ did in NetBSD.
No functional change.
Ingo Schwarze [Thu, 7 Jan 2016 21:03:54 +0000 (21:03 +0000)]
This code wasted memory by allocating sizeof(enum termfont *)
where only sizeof(enum termfont) is needed.
Fixes CID 1288941. From christos@ via wiz@, both at NetBSD.
Ingo Schwarze [Thu, 7 Jan 2016 20:19:01 +0000 (20:19 +0000)]
Recursive "define" was not detected because "lim" was never
incremented, causing infinite loops.
Fixing CID 1288962. From christos@ via wiz@, both at NetBSD.