Updates to better handle RISC-V cross envs and alternate OpenJDK repos

[sxa]

A few changes here - mostly in the interests of usability of our scripts:

• Allow cross compilation in environments other than x86-64 (NATIVE_API_ARCH variable)
• Set BUILD_CC and BUILD_CXX for Bisheng if building natively on RISC-V
• Abort immediately if the build JDK is not usable.
• Allow Debian/Ubuntu-supplied native cross compilers from /usr/bin to be used
• Allow RISCV_SYSROOT to be overriden but still use /opt/fedora28_riscv_root *(what Adopt uses) as the default.
• If the tag detection logic fails e.g. building from a different openjdk repository and --disable-adopt-branch-safety is specified, use 8u000-b00 or jdk-nn-0.0.0+0 to allow building

I would imagine this is all pretty non-controversial, although I'm "open to offers" on the last item here. See this Slack thread for the original discussion - I've chosen here to re-use the existing flag to handle all "I'm building from somewhere funky" situations, but it could be an alternate parameter e.g. an "Override the tag to be this value explicitly" (can be done separately in a future PR of course)

I do also wonder whether we should split out temurin as a separate variant to HotSpot now - the more I think about it in various conversations (particuarly if other people are using our scripts to build HotSpot codebases) the more I think it would keep things simple ... Raised proposal in #2671

THIS IS IN DRAFT AS I WANT TO DO SOME FURTHER CHECKS, BUT REVIEWS/COMMENTS ARE VERY WELCOME IN THE MEANTIME.

Signed-off-by: Stewart X Addison sxa@redhat.com

[karianna]

Nitpicks + one linter failure.

[karianna]

Seems brittle on specific version?

[sxa]

Can you explain that comment please?

[karianna]

It's checking for a very specific version of openssl102 - is there a way to future proof that?

[sxa]

This should be a temporary tactical fix - it's a workaround for the fact that the new --withopenssl=fetched added a few weeks ago doesn't currently work on RISC-V, so the future proofing will be for the OpenJ9 team to make the fetched option work.

[karianna]

LGTM now.

[sxa]

Thanks. I'm still doing a few more tests and will merge later this week

[sxa]

Has been tested with the two RISC-V builds (Thanks @andrew-m-leonard) we have and has not made things any worse, but will enable this to be used in more environments, so taking out of draft.