Actually, saving sources into debuginfo packages never worked properly (or at all?) on mipsel.

The thing is that on MIPS, traditionally, the debug information is stored in a bit different way. For starters, debugedit should handle .debug_info section even if it has MIPS-specific SHT_MIPS_DWARF type:

http://git.altlinux.org/people/iv/packages/?p=rpm-build.git;a=commit;h=1d7d10edd03c2bc6557086eca22a115aa8f9dd00

That, at least, makes rpm-build testsuite pass, but there may be more to it. More testing is surely needed.

Btw, possible, you also need commit like this
http://git.altlinux.org/people/vt/packages/?p=rpm-build.git;a=commitdiff;h=bb0b76e75998641fdb94469e381688bcc74e3c8a

(In reply to Vitaly Chikunov from comment #2)
> Btw, possible, you also need commit like this
> http://git.altlinux.org/people/vt/packages/?p=rpm-build.git;a=commitdiff;
> h=bb0b76e75998641fdb94469e381688bcc74e3c8a

Thank you for pointing this out. I would expect that something like that is needed, and it's easy to imagine what exectly (R_MIPS_32 reolcation for EM_MIPS and EM_MIPS_RS3_LE machines).

But for all the binaries I've seen the relocations for .debug_info section are stored in .rel.debug_info (of type REL) and have value zero, so are "uninteresting":

http://git.altlinux.org/people/iv/packages/?p=rpm-build.git;a=blob;f=tools/debugedit.c;h=051eea1227d1411887aef8a02d8d00ff22b834e6#l543

During this weekend, quite a few packages were build with rpm-build with my patch from comment #1, and "Unhandled relocation" error was not triggered in any of them.

Is there a way to make our current toolchain to produce a relocation against .debug_info section that would be "interesting" for setup_relbuf?

That bug was only triggered on kernel modules objects (ko).

(In reply to Ivan A. Melnikov from comment #3)
> http://git.altlinux.org/people/iv/packages/?p=rpm-build.git;a=blob;f=tools/
> debugedit.c;h=051eea1227d1411887aef8a02d8d00ff22b834e6#l543

Is this commit ready to be merged?

> Is this commit ready to be merged?

Yes. It fixes debuginfo generation on mipsel, and makes the tests pass. rpm-build with this commit applied already works in sisyphus_mipsel and p9_mipsel for quite some time, so it's also well tested.

I don't seem to need another part Vitaly mentioned in comment #2, even to build debuginfo-enabled kernels, but I still don't now why mipsel is different from other platforms here and how exactly. That's the only thing that was holding me from pushing this change into the Sisyphus rpm-build more eagerly.

This is fixed in debugedit upstream. With the transition to separately-packaged debugedit, this is fixed in Sisyphus and p10, too.