[aarch64-port-dev ] JTreg failures in aarch64 jdk7u

[Edward Nevill]

Hi,

I have been looking at the jtreg hotspot failures in aarch64 jdk7. There are a large number of failures because many of the jtreg hotspot tests contain code which is specific to jdk8, even though it is a jdk7 build.

This is because the hotspot sub-repo came originally from jdk8 (hs 25.0-b70) instead of the standard jdk7u hotspot (hs 24.80-b07) and there are many tests which depend on jdk8, either because the tests use jdk8 source features, or because they rely on jdk8 APIs.

For example, all (38) of the mathexact tests are invalid and therefore fail because mathexact does not exist on jdk7.

The following is a summary of the results i get for jdk7 hotspot

Test results: passed: 343; failed: 102; error: 1

For comparison, i tried running the same tests on x86 jdk7 with the following results.

Test results: passed: 311; failed: 134; error: 1

I then took the jtreg test suite from the hs24 build and ran it on both aarch64 and x86 vsns of jdk7 with the following results.

aarch64:

Test results: passed: 279; failed: 11; error: 2

x86:

Test results: passed: 286; failed: 4; error: 2

Here aarch64 still gets more failures than x86 because some of the tests make assumptions about testing hs24 vs hs25 (eg. the behaviour of various -XX: options). However the number of test failures is much smaller.

So, what is the best way to resolve this?

1) Prune the existing aarch64 jtreg hotspot tree to remove any tests which are specific to jdk8 or
2) Generate an exclude list to exclude the above tests but leave them in place in the tree or
3) Replace the existing aarch64 jtreg hotspot tree (from hs25) with the one from hs24 (which gets 11 failures) and then either prune or exclude the much smaller set of tests which depend on some feature of hs24.

I am happy to implement whichever solution as long as the solution works and is satisfactory to everyone.

All the best,
Ed.