Hello Matthias,
On 2019-05-24 05:14, Matthias Apitz wrote:
El día viernes, mayo 24, 2019 a las 11:55:09a. m. +0200, Matthias Apitz
escribió:
I tried to use the JT regression test tool jtreg-4.2-b14.tar.gz the
following way (as explained in its docs:
http://openjdk.java.net/jtreg/runtests.html
$ export JT_HOME=/export/home/sisis/guru/jtreg
$ export JT_JAVA=/usr/local/sisis-pap/jdk1.8.0_31/bin/java
$ export
PRODUCT_HOME=/export/home/sisis/guru/jdk8u212-b03/build/solaris-sparcv9-normal-server-release/images/j2sdk-image
(btw: The script has a bug: the shebang must be modified to '/usr/bin/bash')
$ cd test
$ nohup /usr/sfw/bin/gmake all
I followed the hint in http://openjdk.java.net/jtreg/runtests.html to
use the new built JDK for the testing itself. I set:
$ export
JT_JAVA=/export/home/sisis/guru/jdk8u212-b03/build/solaris-sparcv9-normal-server-release/images/j2sdk-image/bin/java
For JT_JAVA, you generally want the JDK you trust the most, which is
usually the boot jdk.
Then I looked in the test/Makefile and started not 'all', but 'default',
i.e.
$ nohup /usr/sfw/bin/gmake all
Running the 'all' target is certainly not expected to pass. Test
stability in general is something that we put a lot of effort into in
JDK 9 and up, but even so, I would not expect all tests to pass in
current mainline either. We have sub selections of tests that are
considered more or less stable.
'default' means as target: jdk_core langtools_jtreg
and these test until now are all passed with 'Execution successful'
Is this good enough as a test?
Building from source, using a different compiler version and a different
environment than has been thoroughly tested is always a risk. We can of
course not say if anything is good enough for you, that is ultimately
your responsibility to figure out.
That said, I don't know what a good selection would be. I do think you
should include something from hotspot as well as that does not seem to
be included in the default target. Severin is putting together some
basic set of tests for jdk8u here [1] so perhaps you could look into his
choices there. There is unfortunately no baseline test results available
that I know of, so there really is no way to know if any test failure
you see is significant.
Ultimately, what should matter is what you need to run, so I would
recommend testing the software you intend to use this JDK with.
[1] http://mail.openjdk.java.net/pipermail/build-dev/2019-May/025585.html
/Erik
