This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: PATCH: Target hooks for exception and eh frame sections


Robert Lipe wrote:
> 
> Stan Shebs wrote:
> 
> > visually inspected for i386-sco3.2v5, m68-nextstep3 (both of which
> > have been broken for awhile - nextstep needs additional patches not
> 
> I've known that EH on OpenServer has been broken in the completely
> obvious, drop-dead way, but I've been awaiting other unrelated breakage
> to settle before tackling this.

I'll bet I'm the first person ever to have tried cross-compiling
from a Mac too... :-)  That _SCO_ELF flag is spooky...

> I'll let Richard speak with authority here, but I think your patch will
> break some i386 systems.  There's "classic COFF", formerly handled
> by ix86_svr3_exception_section and DO_GLOBAL_CTORS_BODY and friends,
> there's OpenServer COFF (arbitrary sections, but with fixed length
> names and _init and _fini that are almost, but not entirely like the
> ELF counterparts, and normal ELF that's used by OpenServer in ELF mode.
> OpenServer is (was) unique in the SVR3 mutants in that it was the only
> system that had both TARGET_ELF and TARGET_COFF so that was used to
> trigger some concessions for that system.

Hmmm, a full set of target hooks could deal with this, but it's
a bit much for the old tangle of macros.

> I think you just blurred the line between the first two with this patch.
> I've been planning to take the remaining OpenServer COFF code behind the
> woodshed and shoot it for some time.  But if there are users of "pure"
> COFF systems (and I suspect there are none remaining) they might not
> like this.  I've been planning to banish TARGET_COFF and care only about
> the ELF mode.  At that time, much of sco5.h goes away and it becomes
> just another ELF target.

I'm OK with backing off on this config, I just wanted to have a squeaky
clean patch.  Given that the time spent on patches and discussion
takes away from time spent improving GCC in other ways, it's probably
worthwhile to think hard about how much time we should expend coming
up with clever ways to support old systems.  (The nextstep code is
another example of severe bitrot that's been sucking up people's time
because it has references to things that are changing.)

Stan


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]