Re: Re: speedup collect2 (by not using it)

Re: Re: speedup collect2 (by not using it)

[John David Anglin]

>In message <200305222040.QAA31262@makai.watson.ibm.com>, David Edelsohn writes:
> >>>>>> Mike Stump writes:
> >
> >> does collect2 ever do anything interesting when 
> >> USE_COLLECT2 isn't defined and -frepo isn't given?  Problem areas might 
> >> be IBM (COLLECT_EXPORT_LIST).  If aix is a problem, we can always do 
> >> collect2 when COLLECT_EXPORT_LIST is defined.
> >
> >      The GCC configuration for AIX always define use_collect2=yes
> >because it needs collect2 to build the list of constructors.
>Similarly for HPUX.

This isn't true anymore under hpux11.  On the 64-bit port, we
don't use collect2.  On the 32-bit port, we no longer use chatr.  We
now use the LD_INIT_SWITCH and LD_FINI_SWITCH macros.  This has the
advantage that constructors/destructors are automatically run when
shared libraries are loaded/unloaded.  This works with dynamic loading
with shl_load.  Unfortunately, we are still stuck with collect2
under hpux 10.x because the linker doesn't support the +init and
+fini options.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

Re: speedup collect2 (by not using it)

[Matt Austern]

On Thursday, May 22, 2003, at 03:27 PM, John David Anglin wrote:

>> In message <200305222040.QAA31262@makai.watson.ibm.com>, David 
>> Edelsohn writes:
>>>>>>>> Mike Stump writes:
>>>
>>>> does collect2 ever do anything interesting when
>>>> USE_COLLECT2 isn't defined and -frepo isn't given?  Problem areas 
>>>> might
>>>> be IBM (COLLECT_EXPORT_LIST).  If aix is a problem, we can always do
>>>> collect2 when COLLECT_EXPORT_LIST is defined.
>>>
>>>      The GCC configuration for AIX always define use_collect2=yes
>>> because it needs collect2 to build the list of constructors.
>> Similarly for HPUX.
>
> This isn't true anymore under hpux11.  On the 64-bit port, we
> don't use collect2.  On the 32-bit port, we no longer use chatr.

But in any case, isn't this a bit tangential to the question Mike
asked?  He asked what happens on systems where USE_COLLECT2
*isn't* defined, and you're talking about systems where it is (or at 
least
was) defined.

			--Matt

Re: speedup collect2 (by not using it)

[John David Anglin]

> But in any case, isn't this a bit tangential to the question Mike
> asked?  He asked what happens on systems where USE_COLLECT2
> *isn't* defined, and you're talking about systems where it is (or at 
> least
> was) defined.

Actually, I think that Mike's patch won't work on hppa64-hp-hpux11*.
It appears that gcc falls back to just using "ld" when it doesn't
run collect2.  This will fail because we have multiple linkers
and at least two supported linkers for the hppa64 port itself.
Thus, gcc will need to use the same linker selection mechanism
as collect2 in determining which linker to run.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

Re: speedup collect2 (by not using it)

[David Edelsohn]

>>>>> Matt Austern writes:

Matt> But in any case, isn't this a bit tangential to the question Mike
Matt> asked?  He asked what happens on systems where USE_COLLECT2
Matt> *isn't* defined, and you're talking about systems where it is (or at 
Matt> least was) defined.

	Mike asked about AIX and systems with ctor requirements like AIX.
Those system define use_collect2=yes.

David