Issue with not detecting gcc when ccache is enabled in the adoptopenjdk+openj9 builds

[sxa]

Ref eclipse-openj9/openj9#1919 (comment) which shows where this is set up.

It appears that if the C compiler (although in this case it's ccache) is set to have environment variables preceding it then the OMR component then configure gets confused. Haven't checked the logic but it could be that openj9 are only setting it up like that on s390x:

Failing job on Linux/s390x: https://ci.adoptopenjdk.net/job/openjdk10_openj9_build_s390x_linux/58/consoleFull
Successful job on Linux/x86-64 where compiler is showing as cc: https://ci.adoptopenjdk.net/job/openjdk10_openj9_build_x86-64_linux/54/consoleFull

[sxa]

./make/autoconf/generated-configure.sh and ./make/autoconf/build-performance.m4 in the main openjdk repository both appear to add the extra variables to the start of the CCACHE

[smlambert]

I'd like to raise the priority of this issue, is someone working on it presently?

[DanHeidinga]

@keithc-ca Are you handling this as part of eclipse-openj9/openj9#1919?

[keithc-ca]

No, I expect this is unrelated to that, but I'd be happy to be wrong.

[DanHeidinga]

Thanks Keith for checking.

@dnakamura Can you take a look at this?

[dnakamura]

This root issue is that dash (default /bin/sh on ubuntu), does not properly handle setting variables when included from an expansion inside of eval.
Boiled down code:

CC="var=1 true"
ac_compiler=\$CC
eval "$ac_compiler"

^ works in bash but not dash. Currently working on a fix

[sxa]

Fix doesn't appear to be resolving the problem based on the last attempt to build with the fix

[irinarada]

@dnakamura - any new updates - this is blocking the 090 binaries testing.

[dnakamura]

Opened a PR for JDK9. working on porting to other versions now

[sxa]

Any update @dnakamura ? It's JDK10 that's first in our priority list.

[keithc-ca]

Devin's new fix is merged for JDK10: ibmruntimes/openj9-openjdk-jdk10#44.

[sxa]

Thanks @pshipton - looks like you pulled this across to openj9-0.9 yesterday :-) Will kick off a build and close this off if successful.

[sxa]

Looks happy on the latest build 100 at https://ci.adoptopenjdk.net/job/openjdk10_openj9_build_s390x_linux :-)

[DanHeidinga]

@sxa555 The change hasn't been applied to JDK8 but the build appears to be passing at Adopt. Are there any other JDK levels that you are aware of that need this change?

[sxa]

@DanHeidinga Yeah 8 was never a problem. 9 is though - we haven't built that off the openj9-0.9 branch yet (assuming there is one) but the latest build off the main openj9 branch in there passed ok, so I guess the fix was applied there too.

[pshipton]

The fix is applied to jdk9 and the openj9-0.9 branch of jdk9, which does exist.

[sxa]

Seeing something that looks very similar to this on my zlinux build with JDK11 at the moment

[keithc-ca]

ibmruntimes/openj9-openjdk-jdk11#9 should fix JDK11.

[pshipton]

ibmruntimes/openj9-openjdk-jdk11#9 is merged.