This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Generate dependencies from $(native-compile)
On Fri, 27 Apr 2012, Roland McGrath wrote:
> Meh. As per previous discussion, I think we are/should be going in
> the direction of eliminating the existing uses of native-compile.
> So there's no sense in making speculative tweaks to it that don't
> actually do anything yet.
Well, the point is to get in cross-rpcgen - overall aim, to end up with
the same files installed from native and cross builds - for which the
current patch version is
<http://sourceware.org/ml/libc-alpha/2012-04/msg01063.html>. And adding
some flags to what goes in $(native-compile) is in some way separable from
splitting a common part of $(native-compile) out of the $(native-compile)
definition and using it for cross-rpcgen.
Maybe if we aim to eliminate $(native-compile) that means there's even
less case for trying to share much between cross-rpcgen and
$(native-compile), and indeed $(compile-mkdep-flags) should actually go in
sunrpc/Makefile on the compilation command there (near the addition to
extra-objs which causes the dependency files to be used)?
> I can take on turning the elf/Makefile cases into scripting rather
> than programs. Maybe someone else wants to grok annexc, isomac, and
> conformtest enough to consider consolidating them all into the latter.
All this ought to be independent of getting cross-rpcgen in. We really
need to avoid unnecessary dependencies between getting in changes in
different areas.
Merging in isomac is more or less a matter of checking that the ISO
definitions in conformtest do actually correspond accurately to ISO C90
and that all headers tested by isomac are properly tested by conformtest.
(There's the complication of isomac checking C94 headers and conformtest
not doing any C94 testing; conformtest should have a C94 mode added, but
it will be the same as C90 mode except for a few headers.)
isomac is expected to pass. conformtest is effectively XFAILed. Maybe we
should move the testing loop from run-conformtest.sh to the makefile, so
that each standard's tests are run separately by the makefile (using
separate temporary directories, obviously) with expectations of whether
they pass or fail also controlled separately - ISO and ISO99 expected to
pass. That would certainly help speed things up for people testing in
parallel, since conformtest is pretty slow.
annexc is expected to fail. It's more sophisticated than isomac in that
it records what's expected from each header, whereas isomac just has a
global list of symbols. It's also more of a major effort to add to
conformtest since it tests the original POSIX.1 (_POSIX_SOURCE) which is
not currently one of the modes supported by conformtest. Although the
extensions (POSIX.2, POSIX.1b and POSIX.1c) up to the version
conformtest.pl treats as POSIX were reasonably well-defined discrete sets,
so maybe marking them appropriately in the expectations isn't too hard for
someone with copies of the relevant documents (I have POSIX.2 and the 1996
edition of POSIX.1 but not the older POSIX.1 documents).
--
Joseph S. Myers
joseph@codesourcery.com