[prev in list] [next in list] [prev in thread] [next in thread] 

List:       mandoc-discuss
Subject:    Re: Segmentation fault on some man pages
From:       Ingo Schwarze <schwarze () usta ! de>
Date:       2019-12-02 21:05:00
Message-ID: 20191202210500.GA91782 () athene ! usta ! de
[Download RAW message or body]

Hello Milan,

Milan P. Stanic wrote on Sat, Nov 30, 2019 at 08:39:26PM +0100:

> Sorry if this is not right list to post bug report

discuss@, tech@, or devel@ are all fine for bug reports.

 - discuss@ is best when you think that many people may be affected
   and/or if it feels important or serious
 - tech@ is better for minor quirks that fewer people are likely
   to be interested in
 - devel@ provides as choice to report a bug privately if you
   prefer that for whatever reason

> but I can't find where to post it.

https://mandoc.bsd.lv/contact.html

> On Alpine linux (musl libc) we discovered 'Segmentation fault' when
> running 'man nft' (nftables man page).

Thanks for reporting.

> We have mandoc (named mdocml=1.14.5-r0 in our distribution) and it works
> fine, except this for 'nft' man page.
> 
> This is excerpt from 'ltrace man nft':
> putwchar(9472, 9472, 0x7fe813c0bbab, 0)
> strcmp(nil, "\\^" <no return ...>
> --- SIGSEGV (Segmentation fault) ---
> +++ killed by SIGSEGV +++

I tried to reporduce with the input file

  https://manpages.debian.org/testing/nftables/nft.8.en.gz

but mandoc-current on OpenBSD-current does not crash.
Is that the exact version of the input file nft.8 that you are using?
If not, could you send a copy of the nft.8 that crashes for you
to me directly (not needed on the list)?

> I looked in source and found problem in tbl_term.c file.
> I made crude hack by removing few lines in tbl_term.c and build mandoc
> again and with these changes it works without (visible to me) problem
> for man pages I tried and for nft man page.
> 
> I'm attaching patch I made, but I think it is not proper fix and hope
> that some of you experienced in this will make proper fix.
> File I'm attaching is just hint.

Hmmm...  The code in mandoc-current does not match the code
on the "-"-lines of your patch below.  Mandoc-current contains:

	/* Print the horizontal line inside this column. */

	lw = cpp == NULL || cpn == NULL ||
	    (cpn->pos != TBL_CELL_DOWN &&
	     (dpn == NULL || dpn->string == NULL ||
	      strcmp(dpn->string, "\\^") != 0))
	    ? hw : 0;
[...]
	/* The horizontal line inside the next column. */

	rw = cpp == NULL || cpn == NULL ||
	    (cpn->pos != TBL_CELL_DOWN &&
	     (dpn == NULL || dpn->string == NULL ||
	      strcmp(dpn->string, "\\^") != 0))
	    ? hw : 0;

I suspect the following commit might already have fixed your problem
about eight months ago:

  https://cvsweb.bsd.lv/mandoc/tbl_term.c#rev1.70
  https://cvsweb.bsd.lv/mandoc/tbl_term.c.diff?r1=1.69&r2=1.70

Does that fix the issue for you?

If so, consider adding a combined patch containing
rev.s 1.70 and 1.71 to your port until i roll a new release:

  https://cvsweb.bsd.lv/mandoc/tbl_term.c.diff?r1=1.69&r2=1.71

Yours,
  Ingo


> --- a/tbl_term.c	2019-03-10 10:56:43.000000000 +0100
> +++ b/tbl_term.c	2019-11-30 14:38:12.918969863 +0100
> @@ -625,8 +626,7 @@
>  		/* Print the horizontal line inside this column. */
>  
>  		lw = cpp == NULL || cpn == NULL ||
> -		    (cpn->pos != TBL_CELL_DOWN &&
> -		     (dpn == NULL || strcmp(dpn->string, "\\^") != 0))
> +		    cpn->pos != TBL_CELL_DOWN
>  		    ? hw : 0;
>  		tbl_direct_border(tp, BHORIZ * lw,
>  		    col->width + col->spacing / 2);
> @@ -669,8 +669,7 @@
>  		/* The horizontal line inside the next column. */
>  
>  		rw = cpp == NULL || cpn == NULL ||
> -		    (cpn->pos != TBL_CELL_DOWN &&
> -		     (dpn == NULL || strcmp(dpn->string, "\\^") != 0))
> +		    cpn->pos != TBL_CELL_DOWN
>  		    ? hw : 0;
>  
>  		/* The line crossing at the end of this column. */
--
 To unsubscribe send an email to discuss+unsubscribe@mandoc.bsd.lv

[prev in list] [next in list] [prev in thread] [next in thread]