Not a QA issue. Reassigning. Looking at iproute2-5.0.0 I would say it's not a toolchain specific bug.
A problem is in the ebuild itself. Not how sed call does not escape ${CFLAGS}:
src_prepare() {
....
sed -i \
-e '/^CC :\?=/d' \
-e "/^LIBDIR/s:=.*:=/$(get_libdir):" \
-e "s:-O2:${CFLAGS} ${CPPFLAGS}:" \
-e "/^HOSTCC/s:=.*:= $(tc-getBUILD_CC):" \
-e "/^DBM_INCLUDE/s:=.*:=${T}:" \
Makefile || die
....
}
ulm@, why it's not a QA issue?
> so far, I've found: > > app-arch/p7zip-16.02-r4 Filed bug #685166 > sys-apps/iproute2-5.0.0 Filed bug #685168 > dev-qt/qtcore-5.12.3 > dev-qt/qtnetwork-5.12.3 > dev-qt/qtsql-5.12.3 > dev-qt/qttest-5.12.3 > dev-qt/qtdbus-5.12.3 > dev-qt/qtconcurrent-5.12.3 > dev-qt/qtgui-5.12.3-r1 > dev-qt/qtwidgets-5.12.3 > dev-qt/qtprintsupport-5.12.3 Filed bug #685170 Please file new failures as individual bugs blocked against this bug. For those who are here for a hint to fix it using '|' sounds like a safer delimiter.
Example iproute the fix would be:
@@ -64 +64 @@ src_prepare() {
- -e "s:-O2:${CFLAGS} ${CPPFLAGS}:" \
+ -e "s|-O2|${CFLAGS} ${CPPFLAGS}|" \
In some cases a better fix would be to stop putting flags inline and instead adding variables to Makefiles. (In reply to Sergei Trofimovich from comment #2) > ulm@, why it's not a QA issue? This bug is blocking the gcc-9 tracker, so IMHO it should be assigned to toolchain. Also, using : as a separator in sed is not a QA issue per se. (In reply to Ulrich Müller from comment #6) > (In reply to Sergei Trofimovich from comment #2) > > ulm@, why it's not a QA issue? > > This bug is blocking the gcc-9 tracker, so IMHO it should be assigned to > toolchain. gcc-9 is not the first version that has options with colons. I dropped it from blocker. > Also, using : as a separator in sed is not a QA issue per se. The bug is about the ebuild failure in src_prepare() that is relatively easy to find. If qa@ does not plan to expand the tools or documentation to detect that kind of breakage it's fine by me. If someone wants a few more commits into ::gentoo there are a lot more probken packages in the tree like that:
$ git grep 's:' | fgrep CFLAGS | fgrep sed | wc -l
130
A few selected examples:
$ git grep 's:' | fgrep CFLAGS | fgrep sed
...
app-cdr/xbiso/xbiso-0.6.1-r2.ebuild: sed -i -e 's:C) $(CFLAGS):C) $(LDFLAGS) $(CFLAGS):' Makefile.in || die #337769
app-cdr/xbiso/xbiso-0.6.1-r3.ebuild: sed -i -e 's:C) $(CFLAGS):C) $(LDFLAGS) $(CFLAGS):' Makefile.in || die #337769
app-crypt/md4sum/md4sum-0.02.03-r2.ebuild: sed -i -e "s:CFLAGS=:CFLAGS=${CFLAGS} :g" \
app-crypt/seahorse-sharing/seahorse-sharing-3.8.0_p20151117.ebuild: sed -e 's:$CFLAGS -g -O0:$CFLAGS:' \
app-editors/elvis/elvis-2.2.0-r6.ebuild: sed -i -e "s:gcc -O2:$(tc-getCC) ${CFLAGS}:" Makefile || die "sed 1 failed"
app-editors/elvis/elvis-2.2.0-r7.ebuild: sed -i -e "s:gcc -O2:$(tc-getCC) ${CFLAGS}:" Makefile || die "sed 1 failed"
...
(In reply to Michał Górny from comment #5) > In some cases a better fix would be to stop putting flags inline and instead > adding variables to Makefiles. In almost all cases this is the right thing to do. Instead of sed'ing the contents of the $CFLAGS variable into the target, the variable reference itself should be inserted into the target (if the target is shell, or a Makefile, or something that understands variables). In other words, instead of creating a file that does $(CC) -march=native -O2 -pipe ... we should be creating one that does $(CC) $(CFLAGS) ... The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/devmanual.git/commit/?id=0b00ff1312e734ae453a9f61f504bdc400b2b2df commit 0b00ff1312e734ae453a9f61f504bdc400b2b2df Author: Sam James <sam@gentoo.org> AuthorDate: 2021-03-20 14:48:24 +0000 Commit: Ulrich Müller <ulm@gentoo.org> CommitDate: 2021-03-29 20:41:37 +0000 tools-reference/sed: emphasise need for delimiter care Bug: https://bugs.gentoo.org/685160 Signed-off-by: Sam James <sam@gentoo.org> Signed-off-by: Ulrich Müller <ulm@gentoo.org> tools-reference/sed/text.xml | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) |
with gcc 9.1 one can now use -falign-functions=32:25:16 to say align to a 32 byte boundary, unless you move more than 24 bytes, then align to a 16 byte boundary. same for -falign-jumps, -falign-labels, -falign-loops. adding that to CFLAGS, however, causes emerging a number of packages to fail during src_prepare because they have sed oneliners that use : to delimit substitution: >>> Preparing source in /var/tmp/portage/sys-apps/iproute2-5.0.0/work/iproute2-5.0.0 ... * Applying iproute2-3.1.0-mtu.patch ... [ ok ] * Applying iproute2-4.20.0-configure-nomagic.patch ... [ ok ] sed: -e expression #3, char 152: unknown option to `s' * ERROR: sys-apps/iproute2-5.0.0::gentoo failed (prepare phase): * (no error message) * * Call stack: * ebuild.sh, line 124: Called src_prepare * environment, line 3506: Called die * The specific snippet of code: * sed -i -e '/^CC :\?=/d' -e "/^LIBDIR/s:=.*:=/$(get_libdir):" -e "s:-O2:${CFLAGS} ${CPPFLAGS}:" -e "/^HOSTCC/s:=.*:= $(tc-getBUILD_CC):" -e "/^DBM_INCLUDE/s:=.*:=${T}:" Makefile || die; so far, I've found: app-arch/p7zip-16.02-r4 sys-apps/iproute2-5.0.0 dev-qt/qtcore-5.12.3 dev-qt/qtnetwork-5.12.3 dev-qt/qtsql-5.12.3 dev-qt/qttest-5.12.3 dev-qt/qtdbus-5.12.3 dev-qt/qtconcurrent-5.12.3 dev-qt/qtgui-5.12.3-r1 dev-qt/qtwidgets-5.12.3 dev-qt/qtprintsupport-5.12.3