Re: segfault for git log --graph --no-walk --grep a

[Jeff King <peff@peff.net>]

On Mon, Feb 11, 2013 at 12:36:52PM -0800, Junio C Hamano wrote:

> Junio C Hamano <gitster@pobox.com> writes:
> 
> > Jeff King <peff@peff.net> writes:
> >
> >> On Fri, Feb 08, 2013 at 04:47:01PM -0800, Junio C Hamano wrote:
> >>
> >>> > Yeah, that actually is a good point.  We should be using logmsg_reencode
> >>> > so that we look for strings in the user's encoding.
> >>> 
> >>> Perhaps like this.  Just like the previous one (which should be
> >>> discarded), this makes the function always use the temporary strbuf,
> >>> so doing this upfront actually loses more code than it adds ;-)
> >>
> >> I didn't see this in What's Cooking or pu. We should probably pick an
> >> approach and go with it.
> >>
> >> I think using logmsg_reencode makes sense. I'd be in favor of avoiding
> >> the extra copy in the common case, so something like the patch below. If
> >> you feel strongly about the code cleanup at the minor run-time expense,
> >> I won't argue too much, though.
> >
> > Sounds good to me.  Care to do the log message while at it?
> 
> Heh, how about this?  I still need a sign-off from you.

I'm working on the log message and tests right now. There's also a minor
code fixup needed to compile with -Wall.

>     log --grep: look for the given string in log output encoding
>     
>     We used to grep in the raw commit buffer contents, possibly pieces
>     of notes encoded in log output encoding appended to it, which was
>     insane.
>     
>     Convert the contents of the commit message also to log output
>     encoding before looking for the string.  This incidentally fixes a
>     possible NULL dereference that can happen when commit->buffer has
>     already been freed, which can happen with
>     
>     	git commit -m 'text1' --allow-empty
>     	git commit -m 'text2' --allow-empty
>     	git log --graph --no-walk --grep 'text2'
>     
>     which arguably does not make any sense (--graph inherently wants a
>     connected history, and by --no-walk the command line is telling us
>     to show discrete points in history without connectivity), and we
>     probably should forbid the combination, but that is a separate
>     issue.

I'll use bits of that. I had sort of punted on the "how to reproduce the
segfault" issue entirely because you had noted that it was not a sane
thing to do. Still, I think it makes sense to mention it with the caveat
you give here.

-Peff