Re: makewhatis SEGV

On Fri, 14 Aug 2026 13:47:05 +0200,
Ingo Schwarze <[email protected]> wrote:
> 
> Hi Thomas,
> 
> Thomas de Grivel wrote on Fri, Aug 14, 2026 at 12:14:27PM +0200:
> - On Fri, Aug 14, 2026, 11:46 Kirill A. Korinsky <[email protected]> wrote:
> !! On Fri, 14 Aug 2026 06:56:27 +0200, Thomas de Grivel wrote:
> --- On Thu, Aug 13, 2026 at 10:50 Stuart Henderson wrote:
> !!!! On 13 August 2026 09:24:56 Thomas de Grivel <[email protected]> wrote:
> 
> !!!!! # pkg_add -u
> !!!!! [...install ok...]
> !!!!! makewhatis([...a lot of manpage paths...]) failed: killed by signal
> !!!!! SEGV (core dumped)
> 
> ---- Without either a backtrace or the actual output, there's nothing
> ---- anyone can do with this report . Information about your system (openbsd
> ---- version, arch, etc) missing too.
> 
> !!! OpenBSD snapshot from yesterday, amd64.
> 
> -- Have you got installed erlang?
> 
> ! Yes
> 
> I also have erlang installed and can run pkg_add -u and makewhatis
> without problems, so the information about erlang is either
> irrelevant or insufficient.
> 
> Can you reproduce a crash by running just
> 
>   $ makewhatis
> 
> If that segfaults, then please try
> 
>   $ cd /usr/src/usr.bin/mandoc/
>   $ cvs up -dP
>   $ rm -f obj
>   $ make clean
>   $ make obj
>   $ make clean
>   $ make DEBUG='-g -O0'
>   $ cd obj
>   $ ln -s mandoc makewhatis
>   $ egdb ./makewhatis
>   (gdb) run
> 
> Then, at the point where it SEGVs, in addition to the full backtrace,
> it would be useful to know which file is being parsed at the time of
> the crash, by inspecting the relevant data in relevant high-level
> functions, including, but not necessarily limited to, the variable
> *filename in the function mparse_readfd(), file mandoc/read.c.
> 
> If a vanilla "makewhatis" with no arguments does not SEGV on your
> system, tracking it down becomes even harder.  In that case,
> triggering the bug likely requires parsing specific files in a
> specific order, and the crash is likely due to information leaking
> from one parse sequence to the next (which in and of itself is often
> already a bug), and then causing trouble by being bogus information
> in the new parsing context.
> 

I had saw it a few times and had reported to volker@ (Cc-ed) directly.

As far as I recall he said that some working on this, but I haven't recall who.

-- 
wbr, Kirill