This is the mail archive of the
mailing list for the glibc project.
Re: [COMMITTED PATCH] Consolidate not-cancel.h files.
- From: Roland McGrath <roland at hack dot frob dot com>
- To: "Joseph S. Myers" <joseph at codesourcery dot com>
- Cc: "GNU C. Library" <libc-alpha at sourceware dot org>
- Date: Wed, 14 May 2014 14:21:40 -0700 (PDT)
- Subject: Re: [COMMITTED PATCH] Consolidate not-cancel.h files.
- Authentication-results: sourceware.org; auth=none
- References: <20140514180923 dot D0EBB2C39C7 at topped-with-meat dot com> <Pine dot LNX dot 4 dot 64 dot 1405142059270 dot 21615 at digraph dot polyomino dot org dot uk>
> On Wed, 14 May 2014, Roland McGrath wrote:
> > There are in fact only two versions of not-cancel.h actually in use: the
> > sysdeps/generic one for non-Linux, and the NPTL one living in
> > nptl/sysdeps/unix/sysv/linux/i386/. The diff looks nontrivial only because
> > moving nptl/sysdeps/unix/sysv/linux/i386/not-cancel.h verbatim to
> > sysdeps/unix/sysv/linux/ clobbered the unused file that was there before.
> I don't believe that file was unused; plenty of architectures didn't have
> their own not-cancel.h files and so used the sysdeps/unix/sysv/linux/ one.
> It may well be that the differences were *unnecessary*; I haven't thought
> what what this file should optimally look like.
You're right. I misread the set of files. Sorry about that.
The machines that were using the old sysdeps/unix/sysv/linux/not-cancel.h
The difference between the old and new files is that the old one used
INLINE_SYSCALL for everything, while the new one uses the __*_nocancel
entry points that are defined along with the ABI entry points by the
I think we want uniformity across machines between those two choices,
though conceivably there is a different optimization choice to be made on
some machines than others. My inclination is to leave things as they are
now and just have everybody test that nothing broke. Does that sound OK?