This is the mail archive of the
mailing list for the GDB project.
Re: RFA: always default to using the libiberty regex
On Tue, Jan 28, 2003 at 09:35:33AM -0500, Andrew Cagney wrote:
> >On Mon, Jan 27, 2003 at 04:56:56PM -0800, Martin M. Hunt wrote:
> >>This patch deletes the configure code to check the OS implementation of
> >>regex and default to that. The default will now always be the builtin.
> >>2003-01-27 Martin M. Hunt <email@example.com>
> >> * configure.in: Revert check for system regex. Use builtin regex by
> >> * configure: Rebuilt.
> >I'm still not convinced this is a good idea.
> >Context: it's a bug in the system's GNU C library, and should be fixed
> >as such. All the rest of us who have a version of glibc which has this
> >issue addressed don't have a problem, and I don't really want to carry
> >around yet another statically linked copy of regex if I don't need to.
> >Since it doesn't manifest on my system, I suspect it is fixed in glibc
> >2.3.1; it's another piece of fallout from Red Hat's choice of using the
> >brand-new barely-tested glibc 2.2.93 for their desktop product.
> There are several choices here:
> - GDB prefers the installed regex.
> This is what the current config is doing.
> I think it is telling that the patch hasn't even been in for a month
> and, already, we've hit problems. No matter where the problem is, it is
> GDB that will get the blame. I also believe that way back when this was
> last debated, specific regex releases were identified as problematic.
> - GDB prefers the bundled regex.
> This is what the old config was doing.
> Doing this means that GDB needs to, at regular intervals, upgrade that
> code. We do this now for readline. It also means that the GDB
> developers are insulated from problems with the utility libraries.
> As best I can tell, the only people that really benefit from the system
> regex are those making distros - it saves them the hassle of having to
> remember --with-..regex=. GDB developers don't benefit - we all need to
> specify the bundled regex as otherwize we can't reproduce each others
> results (cf Martin's problem). GDB users don't benefit as they now get
> a GDB of unknown quality.
Put that way, the patch is OK with me.
MontaVista Software Debian GNU/Linux Developer