rpms/git-core/devel .cvsignore, 1.10, 1.11 git-core.spec, 1.9, 1.10 sources, 1.10, 1.11

[Chris Wright]

* Ralf Corsepius (rc040203 at freenet.de) wrote:
> On Mon, 2005-10-31 at 14:35 -0500, Warren Togami wrote:
> > Chris Wright wrote:
> > > 
> > > It builds fine, and has internal macros to handle zlib-devel < 1.2.
> > > It's runtime that fails (apparently, I haven't recreated this myself,
> > > it was reported by someone running older distro).  But git-core uses
> > > zlib's *Bound functions which seem to have been introduced in the
> > > .so w/out bumping any library version (between zlib 1.1 and zlib 1.2).
> > > So RPM creates implicit requirement on libz.so.1, which is satisified on
> > > machine that has either zlib 1.1 or zlib 1.2.  However will fail during
> > > runtime with the former.  So it seems that zlib >= 1.2 is best way to
> > > handle that.  What do you think?
> 
> So, the purpose of this requirement is to force users running zlib-1.1
> to upgrade to zlib-1.2? If users are pulling zlib and git-core rpms from
> the same site (FC rsp. FE), this should not be a issue. All versions of
> zlib having ever been shipped with FC had been > 1.2.

At least to let them know that the package won't work.  A clue to
rebuild from src.rpm.

> Nevertheless, you still will want:
> BuildRequires: zlib-devel >= 1.2
> 
> Then, the package won't build on distros with older zlibs and will
> prevent those people building against insufficient libz's to trip over
> this issue, later (linking should fail).

Building with zlib < 1.2 is handled by the package (doesn't use the new
*Bound functions), it's build on zlib 1.2, install on zlib 1.1 that fails,
because zlib didn't change soname.  This is possible if you build your
own rpm w/out FE build system to add _fc? to release.  Patch came from
upstream where user hit this situation.

> > Usually we advise against adding requirements like this, because it is 
> > unsupported to run packages built on one distro on another.  However in 
> > a few cases like this were the soname was not bumped, I suppose it is 
> > fine.
> Well, this "R: zlib >= 1.2" shouldn't do any harm ;)
> 
> >   Please leave a comment above it explaining why it is that way though.
> If you want to keep it, yes, but I don't see much reason for keeping the
> equation.

My interest is in keeping deviation from upstream to a minimum.

thanks,
-chris