This is the mail archive of the
libc-hacker@cygnus.com
mailing list for the glibc project.
Re: Please compile glibc 2 with egcs 1.1.1/Linux
- To: hjl@lucon.org (H.J. Lu)
- Subject: Re: Please compile glibc 2 with egcs 1.1.1/Linux
- From: Zack Weinberg <zack@rabi.columbia.edu>
- Date: Thu, 21 Jan 1999 12:50:14 -0500
- cc: geoffk@ozemail.com.au (Geoff Keating), libc-hacker@cygnus.com
On Thu, 21 Jan 1999 07:56:15 -0800 (PST), H.J. Lu wrote:
>>
>> > From: hjl@lucon.org (H.J. Lu)
>> > Date: Wed, 20 Jan 1999 10:28:19 -0800 (PST)
>> > Cc: libc-hacker@cygnus.com
>>
>> > If that is a problem, we should have our own crt*.o for libm.so in glibc
>> > 2.1.
>>
>> How is libm.so different from any other shared library? These
>> low-level details shouldn't affect it specially.
>>
>
>It is because libm.so is a very basic shared library for C++.
>libstdc++.so is linked with it. Zack doesn't want to see any
>references to the C++ exception functions, even if they are
>weak and external.
Actually, on a bit more thought, weak extern refs might be fine. Emphasis
on might. With your patch you get weak extern refs in libc too, right? And
how do the eh functions appear in libstdc++ when you link it against libc
2.1 compiled with your patch?
... Since libm and in fact most of our extra libs don't have anything in
.ctors, we might explore not linking crtbegin.o/crtend.o at all. I think
that's too risky to try this close to release.
zw