This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Illegal instruction (core dumped) on i586
- To: steveo at world dot std dot com
- Subject: Re: Illegal instruction (core dumped) on i586
- From: Joe Buck <jbuck at synopsys dot COM>
- Date: Thu, 5 Aug 99 7:40:21 PDT
- Cc: imarkov at cs dot ucla dot edu, john at feith dot com, gcc at egcs dot cygnus dot com
> I'm inclined to agree on this one. As an example, fetchmail is delivered
> as a .tar.gz file as well as a src.rpm.
fetchmail is a completely independent program. Installing fetchmail
won't change the behavior of other programs, and possibly make them
stop working. Installing a new gcc package with a binary-incompatible
C++ library is another matter, one best left to the folks who put out
the distribution.
> Is there anything else I can say to help stimulate the production of gcc
> in both forms?
Even if we wanted to do it, there's no one that we can order to do the job
(the release manager and principal developers have their hands full and,
to the extent that they have free time, would prefer to improve the
compiler to dealing with packaging issues). It would need to be done by a
skilled, careful volunteer who would not announce the result until he/she
had conducted thorough testing. libstdc++/libc collisions are a frequent
problem on Linux, and this has to be managed carefully when doing binary
distributions.
Even then, the RPM should not install the new compiler in /usr/bin,
wiping out the system compiler. For one thing, people would no longer
be able to build Linux kernels without manual intervention (as one
must specify -fno-strict-aliasing to build the kernel correctly).