On 2015-06-11 08:51, Erik Joelsson wrote:
Hello Omair,
This looks good to me, thanks for catching it!
Looks good to me too.
However, this is not the first time we've encountered a problem like
this. I ran into a similar problem some time ago. It's a very brittle
system we have. I believe we should get rid of the CXXFLAGS and just
have CFLAGS and let them cover c++ compilation as well.
Do we have use cases where C and C++ both are used in the same library,
and requires different flags?
/Magnus
/Erik
On 2015-06-11 00:36, Omair Majid wrote:
Hi,
I discovered a bug in SetupNativeCompilation that results in some C++
code being built with wrong CXXFLAGS.
Bug: https://bugs.openjdk.java.net/browse/JDK-8087156
Webrev:
http://cr.openjdk.java.net/~omajid/webrevs/8087156-setupnativecompilation/00/
It turns out that debugging flags are automatically added to the
CXXFLAGS which makes the code that copies over CFLAGS to CXXFLAGS if
CXXFLAGS is empty fail. This causes the desired CXXFLAGS to not be set
at all.
Depending on the code, these CXXFLAGS can include -DPRODUCT. In other
words, some JDK code is currently being built as if in non-product mode.
An example of this is libunpack:
jdk/make/lib/Lib-jdk.pack200.gmk:
$(eval $(call SetupNativeCompilation,BUILD_LIBUNPACK, \
... snip ...
CFLAGS_release := -DPRODUCT, \
... snip ...
The same issue exists in OpenJDK 8.
Thanks,
Omair
