Ports 104877 causing big problems

[Doug Barton]

Ade Lovett wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> 
> On Mar 09, 2007, at 17:30 , Jean-Yves Lefort wrote:
> 
>> On Fri, 9 Mar 2007 17:05:31 -0800
>> Ade Lovett <ade at FreeBSD.org> wrote:
>>> 3. Ports that *are* affected by this issue (assuming the issue still
>>> exists) can be fixed in a more relaxed manner (eg: a conversion of
>>> GNU_CONFIGURE=YES to USE_AUTOTOOLS=configurehack [implying
>>> GNU_CONFIGURE=YES]) than a time-T switch.  It will also allow for
>>> such affected ports to have PORTREVISIONs bumped by the respective
>>> maintainers so as to more clearly identify improved operation to the
>>> consumers of those ports.
>>
>> All ports that use libtool to produce a program or shared library are
>> affected by this issue.
> 
> This in turn implies that in case that there is an issue, and something 
> needs to be done about it, then silently changing the semantics of 
> GNU_CONFIGURE is not an appropriate solution.  In order for the change, 
> should it be required, to be communicated to all the consumers of the 
> FreeBSD ports tree, and not that subset that happens to read esoteric 
> discussions on a high volume mailing list, this requires that we use the 
> tools available to us, ie: bumping PORTREVISION.

On it's face I find the idea of bumping PORTREVISION for every port 
that uses libtool in any form a sort of silly proposition. The change 
in behavior was introduced in Mk/*, I think it's reasonable to expect 
that the fix happen there too.

> 1.  Identify if there is still a problem.

I just performed the following experiment. Please let me know if you 
think this constitutes empirical evidence of a problem.

1. csup'ed to the latest ports tree
2. Saved a copy of /var/db/pkg/libgpg-error-1.4/+REQUIRED_BY
3. pkg_delete'd mtr-0.72, which was listed in the +REQUIRED_BY file
4. Checked the file, mtr was gone.
5. Built and installed mtr
6. mtr is back in /var/db/pkg/libgpg-error-1.4/+REQUIRED_BY

There is no reason that mtr would need libgpg-error, and there is no 
dependency for it in the Makefile. It does however set GNU_CONFIGURE.

> 2b. If yes, add a new stanza to USE_AUTOTOOLS, for simplicities sake 
> we'll call it lthack, which defines GNU_CONFIGURE, and also wanders 
> through the configuration files performing the appropriate hackery.

I don't think this is reasonable, or realistic. It would require that 
every single port which uses GNU_CONFIGURE gets changed.

> 3.  Maintainers that have ports affected by the issue go in to the 
> Makefile, chunk GNU_CONFIGURE, add USE_AUTOTOOLS= lthack, bump 
> PORTREVISION, and move on to the next one.

The burden for fixing problems caused by the infrastructure should not 
be placed on maintainers generally. Specifically, this problem is too 
widespread, too nasty, and too urgently in need of a fix to warrant 
the delays that asking maintainers to make these changes would cause.

If a short term fix that is less elegant needs to happen now so that 
time can be spent on a more elegant one in the longer term, so be it.

Doug

-- 

     This .signature sanitized for your protection