From mboxrd@z Thu Jan 1 00:00:00 1970 X-Msuck: nntp://news.gmane.org/gmane.linux.lib.musl.general/10668 Path: news.gmane.org!.POSTED!not-for-mail From: Rich Felker Newsgroups: gmane.linux.lib.musl.general Subject: Re: [PATCH] configure: check whether linker works too Date: Mon, 24 Oct 2016 19:23:08 -0400 Message-ID: <20161024232308.GE19318@brightrain.aerifal.cx> References: <004a5919-eb25-d4bf-a680-57049fa4d428@gmail.com> Reply-To: musl@lists.openwall.com NNTP-Posting-Host: blaine.gmane.org Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Trace: blaine.gmane.org 1477351431 8940 195.159.176.226 (24 Oct 2016 23:23:51 GMT) X-Complaints-To: usenet@blaine.gmane.org NNTP-Posting-Date: Mon, 24 Oct 2016 23:23:51 +0000 (UTC) User-Agent: Mutt/1.5.21 (2010-09-15) To: musl@lists.openwall.com Original-X-From: musl-return-10681-gllmg-musl=m.gmane.org@lists.openwall.com Tue Oct 25 01:23:47 2016 Return-path: Envelope-to: gllmg-musl@m.gmane.org Original-Received: from mother.openwall.net ([195.42.179.200]) by blaine.gmane.org with smtp (Exim 4.84_2) (envelope-from ) id 1byoaP-0007D0-Pk for gllmg-musl@m.gmane.org; Tue, 25 Oct 2016 01:23:21 +0200 Original-Received: (qmail 19636 invoked by uid 550); 24 Oct 2016 23:23:22 -0000 Mailing-List: contact musl-help@lists.openwall.com; run by ezmlm Precedence: bulk List-Post: List-Help: List-Unsubscribe: List-Subscribe: List-ID: Original-Received: (qmail 19612 invoked from network); 24 Oct 2016 23:23:21 -0000 Content-Disposition: inline In-Reply-To: <004a5919-eb25-d4bf-a680-57049fa4d428@gmail.com> Original-Sender: Rich Felker Xref: news.gmane.org gmane.linux.lib.musl.general:10668 Archived-At: On Mon, Oct 24, 2016 at 06:05:02PM -0500, Laine Gholson wrote: > As a result of http://www.openwall.com/lists/musl/2016/10/22/3 > it has been shown that musl's current compiler sanity checks > are inadequate. This patch checks if the compiler as a whole > works properly, and moves the CPPFLAGS/CFLAGS/LDFLAGS sanity > checks after the 'compiler works' check. > > Patch and patch signature attached > > Thanks, > Laine Gholson > >From 99fd54d6f2b7c84a02c1ca0d5f1397d709c14313 Mon Sep 17 00:00:00 2001 > From: Laine Gholson > Date: Mon, 24 Oct 2016 17:49:58 -0500 > Subject: [PATCH] configure: check whether linker works too > > and move flag sanity checks to a seperate test > --- > configure | 13 +++++++++++-- > 1 file changed, 11 insertions(+), 2 deletions(-) > > diff --git a/configure b/configure > index 707eb12..e7d565c 100755 > --- a/configure > +++ b/configure > @@ -236,8 +236,17 @@ printf "%s\n" "$CC" > test -n "$CC" || { echo "$0: cannot find a C compiler" ; exit 1 ; } > > printf "checking whether C compiler works... " > -echo "typedef int x;" > "$tmpc" > -if output=$($CC $CPPFLAGS $CFLAGS -c -o /dev/null "$tmpc" 2>&1) ; then > +echo 'int main(void) { return 0; }' > "$tmpc" > +if output=$($CC -o /dev/null "$tmpc" 2>&1) ; then > +printf "yes\n" > +else > +printf "no; compiler output follows:\n%s\n" "$output" > +exit 1 > +fi As-is, this is not an acceptable change. When bootstrapping a new toolchain, linking executables does not work because there is no libc to link to. That's why we only test compiling and not linking. The only need for a linker in the build process is building a shared libc.so, and there are some linking tests that use -shared (which can work without a libc.so, as long as -nostdlib is also being used). It would be acceptable to add a linking test on which support for shared library output depends; it should probably cause configure to fail if --enable-shared was used (shared=yes) and otherwise should cause shared=auto to become shared=no. > +printf 'checking whether *FLAGS are sane... ' > +echo 'int main(void) { return 0; }' > "$tmpc" > +if output=$($CC $CFLAGS $CPPFLAGS $LDFLAGS -o /dev/null "$tmpc" 2>&1) ; then > printf "yes\n" > else > printf "no; compiler output follows:\n%s\n" "$output" I don't think this test is needed; you'll get a failure at make time if you included bad flags. If there is a reason to prefer doing such a test, there need to be separate ones for compiling and linking (since linking might not work, e.g. if you have a compiler that can't produce shared libraries). Rich