Hi,
I've just introduced libint2 into Fedora. It is a package that contains a) a compiler for generating the source code for an optimized library b) the generated library itself which is used by end users
I'd like to get branches for the package for EPEL5 and EPEL6 as well, but the problem is that building the compiler needs C++11 support, which is not available in EPEL.
Is it OK if I do step a) locally and use the pregenerated sources for the EPEL branches? The Fedora branches would still do the whole bootstrap process.
On Fri, May 10, 2013 at 11:22:32AM +0300, Susi Lehtola wrote:
Hi,
I've just introduced libint2 into Fedora. It is a package that contains a) a compiler for generating the source code for an optimized library b) the generated library itself which is used by end users
I'd like to get branches for the package for EPEL5 and EPEL6 as well, but the problem is that building the compiler needs C++11 support, which is not available in EPEL.
Is it OK if I do step a) locally and use the pregenerated sources for the EPEL branches? The Fedora branches would still do the whole bootstrap process.
Is this for bootstrap-only? ie: You can build once with the locally generated (a) and then immediately rebuild the package using the compiler in the just-built package? That can probably get an exception from the FPC using the bootstrap guidelines fairly easily.
If the EPEL branches will always have to have step (a) pregenerated then the outlook is a lot more grim but we can look at that if we have to.
-Toshio
On Fri, 10 May 2013 11:01:14 -0700 Toshio Kuratomi a.badger@gmail.com wrote:
On Fri, May 10, 2013 at 11:22:32AM +0300, Susi Lehtola wrote:
Is it OK if I do step a) locally and use the pregenerated sources for the EPEL branches? The Fedora branches would still do the whole bootstrap process.
Is this for bootstrap-only? ie: You can build once with the locally generated (a) and then immediately rebuild the package using the compiler in the just-built package? That can probably get an exception from the FPC using the bootstrap guidelines fairly easily.
If the EPEL branches will always have to have step (a) pregenerated then the outlook is a lot more grim but we can look at that if we have to.
The library is for computing special six-dimensional integrals, for which there are many possible ways to reach the wanted result using recursion relations. The compiler has some heuristics to determine what is the shortest path, and generates the code that implements this path.
So, the compiler isn't bootstrap-only, but given the above the sources generated with the same configure flags should end up to be equivalent. Upstream releases only stable tarballs of the generated sources that are aimed for end users. The compiler doesn't have these kind of stable releases, it's just available in a VCS.
On Fri, 10 May 2013 21:31:14 +0300 Susi Lehtola jussilehtola@fedoraproject.org wrote:
So, the compiler isn't bootstrap-only, but given the above the sources generated with the same configure flags should end up to be equivalent. Upstream releases only stable tarballs of the generated sources that are aimed for end users. The compiler doesn't have these kind of stable releases, it's just available in a VCS.
Now, it would also seem that debuginfo isn't generated properly when the library is built alongside the compiler [1], so it would seem to be preferable also for this reason to use a pregenerated source tarball.
https://bugzilla.redhat.com/show_bug.cgi?id=961963
This would naturally also reduce the size of the build logs, which are now pretty huge (360MB).
On Fri, May 10, 2013 at 09:31:14PM +0300, Susi Lehtola wrote:
On Fri, 10 May 2013 11:01:14 -0700 Toshio Kuratomi a.badger@gmail.com wrote:
On Fri, May 10, 2013 at 11:22:32AM +0300, Susi Lehtola wrote:
Is it OK if I do step a) locally and use the pregenerated sources for the EPEL branches? The Fedora branches would still do the whole bootstrap process.
Is this for bootstrap-only? ie: You can build once with the locally generated (a) and then immediately rebuild the package using the compiler in the just-built package? That can probably get an exception from the FPC using the bootstrap guidelines fairly easily.
If the EPEL branches will always have to have step (a) pregenerated then the outlook is a lot more grim but we can look at that if we have to.
The library is for computing special six-dimensional integrals, for which there are many possible ways to reach the wanted result using recursion relations. The compiler has some heuristics to determine what is the shortest path, and generates the code that implements this path.
Open an FPC request... I'm not too sure I'd vote to approve but I'm only one of the members. probably want to be sure it's clear what is being pregenerated (sounds like one source code format to another).
The reasons I probably wouldn't vote to approve with my current understanding of the situation is that you then can't make upstreamable changes directly to the package. You'd have to make changes to the source on your Fedora system, pregenerate new sources there, then import those into the EPEL package. It also removes auditability from the package. We no longer have a direct link from upstream to the sources on pkgs.fedoraproject.org
-Toshio
packaging@lists.fedoraproject.org