Re: MIPS md_apply_fix()(?) problem.

[cgd at broadcom dot com]

"H . J . Lu" <hjl@lucon.org> writes:
> Please follow
> 
> http://sources.redhat.com/ml/binutils/2001-09/msg00268.html
> 
> We may take about the same problem.

Can't tell.  I'd bet yes, but you didn't actually provide any examples
of objects which had differences that were caused by this bug.  (all
of the lo16/hi16 relocs you had that changed were against .bss, which
was fine; it needs to be a reloc against a symbol, not a section.)

Anyway, eventually you point to:

	http://sources.redhat.com/ml/binutils/2001-09/msg00294.html

Richard S. asked you to back out the patch that improved the delay
slot handling, and see if that made any difference.  AFAICT you never
reported the result of that testing, but that could just be me not
seeing the reply (or the archives not pointing me at it).  I believe
if you backed that patch out of your sources and tested it, you would
have seen breakage.

The equivalent patch, which went in as part of Thiemo's checkin on
2001-10-10 (a couple of hours after i pulled my snapshot 8-(), is
indeed what allows the linux kernel to work again in the FSF binutils.
(Why it went in as a part of a much larger patch containing newabi
crud i really do not know; it was unrelated, and at least the
changelogs should have been spaced to indicate that, but "whatever."
water under the bridge now.)


So, the upshot of all this:

* you probably were talking about the same problem.  But really I
  can't tell since none of the examples you provided (that I could see
  in the followups, etc.,) were definitive.

* it's worked around for the linux kernel (and other code which is
  compiled with -G 0), AFAICT mostly by luck.

* Still a real problem for non-"-G 0" code.


(To be honest, when I actually got to digging into this, it really
wasn't that hard to run to ground...  All you've gotta do is fine some
compilation options that get you mostly the same binaries, and then
read enough diffs so that you see the problem.  It took much more time
to binary search + build binutils revs, once i'd isolated a test case
for the problem.)


> Where were you when we were
> discussing it :-(?

Kona, Hawaii, then Hanalei Bay, Kauai.  (...not the best timing for a
vacation involving lots of air travel, as it turned out...)

I didn't get back until after the discussion was done, and, well, had
more important things on my mind than binutils when i got back.  8-)



Anyway, I'll try to create a testsuite entry that shows the problem
(i.e., currently fails 8-), and submit it today or early next week.

I don't think I have the time or clue to actually tackle the _problem_
right now, though, and I _know_ i don't have time to get said clue...



chris