GCC plugins packaging (for AFL++)

[Richard W.M. Jones]

https://bugzilla.redhat.com/show_bug.cgi?id=2459753

AFL is a popular fuzzing tool that we ship in Fedora as
'american-fuzzy-lop'.  It includes various methods for instrumenting
the binaries, but for this case we're looking just at the GCC plugin
method.

Apparently GCC plugins depend on the exact GCC major.minor.release
version (rather than, say, just the major version as with Clang).  Or
AFL *thinks* it needs to depend on the exact version, although you
could set an environment variable to skip the check, see:

https://github.com/AFLplusplus/AFLplusplus/blob/edaf3ad04d64185cae856c75ca46...

Do GCC plugins really depend on the exact major.minor.release of GCC?

If so we're going to have to encode the GCC NVR into the dependencies
of the package which will cause much pain to everyone.

Rich.

[Florian Weimer]

* Richard W. M. Jones:
> Apparently GCC plugins depend on the exact GCC major.minor.release
> version (rather than, say, just the major version as with Clang).  Or
> AFL *thinks* it needs to depend on the exact version, although you
> could set an environment variable to skip the check, see:
>
> https://github.com/AFLplusplus/AFLplusplus/blob/edaf3ad04d64185cae856c75ca46...
>
> Do GCC plugins really depend on the exact major.minor.release of GCC?

They do.  We rebuild the annobin plugin each time we update GCC, and
reconfigure to use that annobin version if the separately built annobin
plugin was built by a different GCC version (and not the currently
installed one).

Thanks,
Florian

[Richard W.M. Jones]

On Thu, Jun 25, 2026 at 06:21:01PM +0200, Florian Weimer wrote:
> Richard W. M. Jones:
> > Apparently GCC plugins depend on the exact GCC major.minor.release
> > version (rather than, say, just the major version as with Clang).  Or
> > AFL *thinks* it needs to depend on the exact version, although you
> > could set an environment variable to skip the check, see:
> >
> > https://github.com/AFLplusplus/AFLplusplus/blob/edaf3ad04d64185cae856c75ca46...
> >
> > Do GCC plugins really depend on the exact major.minor.release of GCC?
>
> They do.  We rebuild the annobin plugin each time we update GCC, and
> reconfigure to use that annobin version if the separately built annobin
> plugin was built by a different GCC version (and not the currently
> installed one).

Sorry for the late reply here.

I don't have any good ideas here, except that I want to avoid having
to rebuild AFL every time we rebuild GCC.  That's not going to make
anyone happy.

Is there a stable subset of APIs or is there just no hope?

Rich.