Re: texinfo (Re: Owl packages with dangerous "tmp" functions)

[Solar Designer]

>   -  tempbase = mktemp (concat ("txiXXXXXX", "", ""));
> [snip]
>   +  tempdir = concat(tempdir, "txi.XXXXXX", "");
>   +  retries = 0x1000;
>   +  do
>   +    {
>   +      if (mktemp (tempdir) == NULL)
>   +        fatal ("mktemp: %s: %s", tempdir, strerror (errno));
>   +      if (mkdir (tempdir, 0700) == 0) break;
>   +      if (errno != EEXIST || !--retries)
>   +        fatal ("mkdir: %s: %s", tempdir, strerror (errno));
>   +      if ((dot = strrchr (tempdir, '.')))
>   +        strcpy (dot, ".XXXXXX");
>   +    }
>   +  while (1);
> 
> This will break on systems with 8+3 limitations on file names (yes, 

Yes, I actually have a "texindex.exe" on my DOS disk.

> Texinfo supports those systems).  Why did you need the dot?

I didn't really need it, this is just more consistent with other
packages.  I thought that "txiXXXXXX" was already 9 characters, so it
wouldn't matter if it's the DOS name or extension that's truncated?

> Also, isn't it better to use mkstemp, if that's available?

mkstemp creates files, not directories.  There's mkdtemp, but it's
even less portable (added to glibc after 2.1).

>   +  atexit(flush_tempfiles_atexit);
> 
> Can we depend on atexit be available on all supported platforms?

I believe so.

> > That other bug I mentioned is this:
> > 
> > -      sort_in_core (&tempfiles[i], MAX_IN_CORE_SORT, newtemp);
> > +      sort_in_core (tempfiles[i], MAX_IN_CORE_SORT, newtemp);
> 
> ??? Are you sure?  sort_in_core accepts a "char **" as its first argument, 
> while tempfiles[i] is a "char *".

Doesn't appear this way:

void
sort_in_core (infile, total, outfile)
     char *infile;

and I actually had texindex crash reporting some pointers in place of
a filename it couldn't open before I've fixed this.

-- 
/sd