Using glibc's regex [was: --with-included-foo={yes,no} and which foo.h to use?]

[Julian Foad]

Stepan Kasal wrote:
> On Thu, Jul 07, 2005 at 11:29:00AM +0100, Julian Foad wrote:
>
> > >         Big regex.[ch] update from latest glibc CVS libc/posix/.
> >
> > It's not appropriate to use the latest development version of some code.  
> > Use the latest officially released version.  (Maybe that's what you meant, 
> > but it doesn't sound like it.)  If the latest released version isn't 
> > useable, please discuss.  Not using officially released versions of 
> > software nearly always leads to trouble at some time.
>
> In general, you are right.  But not in this case.  The development of
> regex in glibc is not wild, there are just bug fixes.
> And there is no point in missing a few non-risky bug fixes which are only
> in the latest CVS.

I understand what you mean, but the way you are talking sounds scarily close to 
the way a novice talks :-)  No offence meant, of course.  Glibc regex has had 
some new features added to it in the recent past, such as case-insensitive 
matching.  How do you know that no new features are being added this week?

In order to be confident that the changes since the last release are only 
"non-risk bug fixes", we have to talk about a specific version of it, and 
evaluate the changes that exist in that version.  Otherwise another change 
could be made between us talking about the "latest" version and us actually 
grabbing it.

So, for the sake of picking a version, let's talk about the version of glibc at 
2005-07-07T00:00 UTC.  Have you examined all the changes since the latest 
release and determined that they are all non-risky bug fixes, or do you know 
that someone has done so?  If we are going to use unreleased code, then we 
share the responsibility for verifying its correctness, rather as if it were 
our own code.

Even if it is "safe" in that sense, I still feel a bit uncomfortable about 
releasing a version of Grep that has a behaviour different from any released 
version in glibc, but on the other hand I can't identify any specific problems 
with doing so.

> And we plan to do active development of regex in the future on our side;
> that will have to be done relative to latest source.

The fact that patches sent to the glibc project ought to be relative to their 
latest source doesn't constrain us to distribute their latest source.

> And, AFAIK, gawk and sed use latest code from glibc CVS, too.

That precedent is a strong argument if someone can confirm it.

> So I think that _in this case_ we should use the CVS version.

Hmm... If those other projects do, and if we talk about a particular version 
rather than "the" CVS version (at some unspecified time), and if we verify its 
correctness independently, then I would accept that it is OK.

- Julian