https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100017
--- Comment #68 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #57)
> I don't see how the patch could be correct.
> [...]
> So, if CXX_FOR_TARGET starts with /, e.g. /opt/gcc-11.1/bin/g++ , then
> the above will use dirname /opt/gcc-11.1/bin/g++ and set ac_dir to
> /opt/gcc-11.1/bin
> But with the proposed change, $RAW_CXX_FOR_TARGET will be
> /opt/gcc-11.1/bin/g++ -nostdinc++ and dirname will fail miserably and not
> set ac_dir to anything:
> dirname /opt/gcc-11.1/bin/g++ -nostdinc++
> dirname: invalid option -- 'n'
> Try 'dirname --help' for more information.
Jakub's analysis is correct. If the cross compiler is found in PATH then the
patch works OK. The configure output shows:
checking where to find the target c++... pre-installed
checking where to find the target c++ for libstdc++... pre-installed
If you build with something like
CXX_FOR_TARGET=/home/jwakely/gcc/aarch64/bin/aarch64-none-linux-gnu-g++ then
configure shows:
checking where to find the target c++... pre-installed in
/home/jwakely/gcc/aarch64/bin
checking where to find the target c++ for libstdc++... pre-installed in
/home/jwakely/gcc/aarch64/bin
But with the patch that prints an error:
checking where to find the target c++... pre-installed in
/home/jwakely/gcc/aarch64/bin
checking where to find the target c++ for libstdc++... dirname: invalid option
-- 'n'
Try 'dirname --help' for more information.
pre-installed in
I don't think that actually causes a problem though. The build still works. The
dirname command is only used to print the result during configure. It's not
great though.
If we do this instead, then we don't have that problem:
diff --git a/Makefile.in b/Makefile.in
index 79c77fccf0f..ada25b8f11b 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -329,7 +329,7 @@ BASE_TARGET_EXPORTS = \
RAW_CXX_TARGET_EXPORTS = \
$(BASE_TARGET_EXPORTS) \
CXX_FOR_TARGET="$(RAW_CXX_FOR_TARGET)"; export CXX_FOR_TARGET; \
- CXX="$(RAW_CXX_FOR_TARGET) $(XGCC_FLAGS_FOR_TARGET) $$TFLAGS"; export
CXX;
+ CXX="$(RAW_CXX_FOR_TARGET) -nostdinc++ $(XGCC_FLAGS_FOR_TARGET)
$$TFLAGS"; export CXX;
NORMAL_TARGET_EXPORTS = \
$(BASE_TARGET_EXPORTS) \
diff --git a/Makefile.tpl b/Makefile.tpl
index ef58fac2b9a..fc10c7a771d 100644
--- a/Makefile.tpl
+++ b/Makefile.tpl
@@ -332,7 +332,7 @@ BASE_TARGET_EXPORTS = \
RAW_CXX_TARGET_EXPORTS = \
$(BASE_TARGET_EXPORTS) \
CXX_FOR_TARGET="$(RAW_CXX_FOR_TARGET)"; export CXX_FOR_TARGET; \
- CXX="$(RAW_CXX_FOR_TARGET) $(XGCC_FLAGS_FOR_TARGET) $$TFLAGS"; export
CXX;
+ CXX="$(RAW_CXX_FOR_TARGET) -nostdinc++ $(XGCC_FLAGS_FOR_TARGET)
$$TFLAGS"; export CXX;
NORMAL_TARGET_EXPORTS = \
$(BASE_TARGET_EXPORTS) \
I think that will mean we get -nostdinc++ twice for build==host builds, but
that's harmless.
The other reason I'm unsure about adding -nostdinc++ to RAW_CXX_FOR_TARGET is
that it's also used for building other target libs, specifically libsanitizer
and libvtv. I think it's correct for them too.
