Problem with build-id and cycling the free pascal compiler

[Joost van der Sluis]

Sorry for the late response, I've missed your mail.

Op zondag 26-04-2009 om 13:41 uur [tijdzone -0700], schreef Roland
McGrath:
> We should figure out why they are different.  (Off hand I don't think this
> issue should have changed in the build tools since F10.)
> 
> Can you give me two binaries they you think ought to be identical, but
> which had different build IDs generated?

Offcourse. I've placed a .tgz file here:
http://menora.cnoc.nl/extern/DifferentBuildIDsExample.tgz

It contains the source of a hello-world application. (hello.pp) And two
binaries (hello and hello1) which differ but were build from the same
source.

There's also a link-script (link.res) and a shell-script (ppas.sh) which
you can use to link the executables yourself. The necessary object files
are also there.

> I want to get to the bottom of the problem before you change anything.
> 
> But I'll note that if you were to use --build-id=none then you'd have your
> rpm build break in the find-debuginfo.sh stage because of missing a build ID.

I know.

> However, --build-id=0x00000000000000000000 (or any 20 hex digits you choose)
> will hard-code that bogus build-ID during your link stages.  That will make
> your comparisons fine if your binaries are really identical.  Then, the
> find-debuginfo.sh stage will regenerate the build IDs after it edits the
> source file names in the DWARF information.

That's a nice trick. I could use that as a last resort.

> But let's find the real source of the problem and fix it rather than
> working around it.

Thanks for the help,

Joost.