This is the mail archive of the
libc-hacker@sourceware.cygnus.com
mailing list for the glibc project.
Re: glibc: Hurd dynamic linker fixes
- To: tb@MIT.EDU
- Subject: Re: glibc: Hurd dynamic linker fixes
- From: Mark Kettenis <kettenis@phys.uva.nl>
- Date: Tue, 7 Jul 1998 10:23:20 +0200 (MET DST)
- CC: libc-hacker@cygnus.com
- References: <199807061635.MAA20160@steve-dallas.MIT.EDU>
Date: Mon, 6 Jul 1998 12:35:54 -0400
From: tb@MIT.EDU (Thomas Bushnell, n/BSG)
I wrote patches to dl-sysdep.c to support using __read (which is now
called by ld.so). I thought I submitted those to Uli, but I don't see
them in the real source.
The point of those changes was to make the fake file descriptions that
ld.so uses be Hurd I/O ports instead of memory object ports; I think
that's better than having __libc_read do a map. It also makes it
possible to support __libc_write, __fxstat, __lseek.
(Your definition of __libc_read also looks wrong, in that it does not
maintain a file pointer.)
I agree that __libc_read looks wrong. But it works because
__libc_read is only used to read the first page of an executable. It
is never called a second time.
However, using I/O ports would be a better solution since that would
make it possible to support writing debug output to arbitrary files,
and shared library profiling.
If you send me your patches, I can try to integrate them in mine.
Mark