This is the mail archive of the
mailing list for the glibc project.
Assuming binutils and GCC features
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: <libc-alpha at sourceware dot org>
- Date: Wed, 25 Jun 2014 16:37:11 +0000
- Subject: Assuming binutils and GCC features
- Authentication-results: sourceware.org; auth=none
Among the cases where the toplevel configure script has
architecture-specific cases for non-ex-ports architectures, while ex-ports
architectures use configure or preconfigure fragments to keep such code in
sysdeps (the preferred approach) is the setting of libc_commonpagesize and
libc_relro_required for the test for linker -z relro support.
Is there any actual likelihood of this support not working for any
supported architecture and binutils version (we require binutils 2.20 or
later)? The binutils support appears to date from 2004. I think it might
make sense simply to remove the configure test for -z relro support, and
just use -z relro without checking if it's working (we already use it
unconditionally, the question is simply whether configure checks for it
being broken) - like we use any number of features unconditionally without
special checks, rather than moving variable settings to several
Another similar case is the test for assembler CFI directives, which also
involves an architecture-specific parameter (different for SPARC). Again,
I'd think we could assume this unconditionally.
There are lots of other configure tests for binutils and GCC features that
seem similarly long-obsolete and ripe for removal, although not affecting
cleanup of architecture cases in architecture-independent files.
Joseph S. Myers