Re: Applying outstanding patches [was: Release what we've got?]

[Tim Waugh]

On Tue, Jun 14, 2005 at 12:26:06AM +0100, Julian Foad wrote:

> it or that I could not reproduce the problem (as with bug #13161 - missing 
> check on memchr() return val in EGexecute - 
> <https://savannah.gnu.org/bugs/?func=detailitem&item_id=13161> - where the 
> "complete test case" succeeds for me even with unpatched grep-2.5.1).

Yes, with pristine grep-2.5.1a it does not crash, and I'm not entirely
sure why not.  The bug is still there by code inspection, but by
chance it doesn't trigger or something.

> "oi" patch:
>   Is it safe to be changing our copy of "regex.h" like this?  Why the 
>   change from #include "regex.h" to <regex.h> ?  Do these changes to search.c 
> make some of our existing case-folding logic redundant?

You need to hold off on that patch until the bundled copy of regex is
brought up to date.

> "w" patch:
>   I'm a bit uncomfortable with the amount of code added by this patch, but 
>   if it really does fix the corresponding tests (which are already in CVS) 
> then probably OK.

Yes, this really is needed for UTF-8.  It applies after the
egf-speedup patch.  With UTF-8 it's tricky to search backwards in a
string, and just foo[-1] is not sufficient.

> "icolor" patch:
>   This just removes some code, saying that it is "redundant and incorrect". 
>   I suspect that it depends on some other patch to have been applied first - 
> and I think maybe I asked and was told - but the patch issue doesn't say 
> so, and it should.  If I'm wrong, and all the tests that currently pass 
> still pass after applying this, that would be great, but I think I tried 
> that before.  I think this is superceded by patch #3767: Remove two 
> match_icase code paths from prline() in src/grep.c.

I've lost track myself.  I think things like this will get better once
a new release is made, and I can update the Fedora Core package.

> "dfa-optional" patch:
>   This makes Grep's operation depend on an environment variable 
> "GREP_USE_DFA".  We shouldn't do that, at least not without providing some 
> documentation.

By all means remove the dependence on that variable, but it is
important to retain the logic that avoids DFA use when the input is
encoded in UTF-8.

> "egf-speedup" patch:
>   We need to read the mailing list to see what was said about this.  It may 
>   be OK, but it may well have been superceded, and it looks to me like it 
> adds far too much code.

It's okay with me to defer this until after 2.5.2 is released.  At the
moment, I have no time to refactor it.  I will certainly be applying
it to the Fedora Core package.

Tim.
*/

pgp5wpBg3_myl.pgp
Description: PGP signature