Jim Meyering <jim@meyering.net> wrote:
Mark McLoughlin <markmc@redhat.com> wrote:
On Wed, 2008-02-20 at 16:42 +0100, Jim Meyering wrote:
If I do as you suggest and move the safewrite definition into util.h, and make it static inline, then both problems go away.
Make sure to try and build this with gcc-4.3.0 - remember that we un-inlined xstrtol() because of:
internal.h:272: error: inlining failed in call to 'xstrtol_i': call is unlikely and code size would grow
Again, the option is to stop using -Winline, but it's probably about time we figured out some proper way of sharing code between libvirt.so and virsh etc. without adding it to the ABI.
How about a libvirtutil libtool convenience (i.e. static) library?
This does sound like a better idea. I'll prepare a patch to does that, and probably move at least these into it, too:
__virStrToLong_i; __virStrToLong_ull;
Actually, it's not so simple. Using a libtool convenience library would work, but to be useful, it would have to expose the safewrite function to all users of libvirt.la. And, since we seem to want to keep -Winline, we probably don't want "static inline" functions, so here's what I've done: #1 Allow safewrite to be used both by library and non-library code: Subject: [PATCH] Move safewrite and saferead to a separate file. #2 Fix the uses of write that started all this: Subject: [PATCH] Use safewrite in place of write, in many cases. Details in the following patches.