Testsuite gcc.target/i386/pr21291.c failures with -fpic/-fPIC ?

[Bonzini Paolo]

> On the 4.2 and 4.1 branches, I do see the failures.  In terms of your
> involvement with that testcase on branches, you added just the comment and
> then reverted it without ever changing the target selector.
>
> http://gcc.gnu.org/viewcvs/branches/gcc-4_1-branch/gcc/testsuite/gcc.target/i386/pr21291.c
> http://gcc.gnu.org/viewcvs/branches/gcc-4_2-branch/gcc/testsuite/gcc.target/i386/pr21291.c

Yes, the patch was reverted there.  I added the comment for documentation
purposes (it was added on mainline only, but applied to all branches),
but not the target selector because I had not run the testcase in -fpic/-fPIC
mode (and hence I didn't know it was failing on the branch).

All my work was reverted on the branches, so the testcase must have been
failing before last July too. 

> Did your patch actually address the -fpic/-fPIC cases or
> was it for something else? 

It was for reducing reload pressure on asms, in general.  Earlier
attempts to fix the same register pressure bug were not aggressive
enough, so they did not "make it" for -fpic/-fPIC (the asm still
used one register too much, as explained in Richard's).

> Was your patch ever reinstalled on the branches?

No, and probably it won't.

> Should the branches have the target selector backported since
> they fail?
>
> Should the comment be removed from mainline since it passes?

I'd say yes.  We might leave the failures as a sign that on mainline
the bug was fixed on the trunk, because in principle the fix is
backportable (though it was considered unsafe for the FSF tree in
the end, and Jakub later fixed a lot of fallout too).  But I don't
think this is our policy.

Paolo