Create Win32 release builds of openjdk8+openj9

[ChengJin01]

Given that we already provided support on JDK8 32bit + OpenJ9 at https://github.com/eclipse/openj9/blob/master/buildenv/Build_Instructions_V8.md#windows, we'd like to get the JDK8 builds (Windows 32bit) built and tested at AdoptOpenJDK for the next OpenJ9 release.

[DanHeidinga]

@ChengJin01 Was there additional machine configuration required to build the 32 bit JDK? Can you highlight any differences here?

[ChengJin01]

Here's a few steps that need to be set up correctly on 32bit:

1. Enure the freetype version supports win 32-bit application (e.g. freetype-2.4.7 which includes required files at the win32 directory)

2. To build the Freetype dynamic and static libraries, open the Visual Studio Command Prompt (VS2010) (see C:\ProgramData\Microsoft\Windows\Start Menu\Programs\Microsoft Visual Studio 2010\Visual Studio Tools) and run:

cd c:\temp
msbuild.exe C:/temp/freetype/builds/win32/vc2010/freetype.vcxproj /p:PlatformToolset=v100 /p:Configuration="Release Multithreaded" /p:PlatformTarget=x86 /p:ConfigurationType=DynamicLibrary /p:TargetName=freetype /p:OutDir="C:/temp/freetype/lib32/" /p:IntDir="C:/temp/freetype/obj32/" > freetype.log
msbuild.exe C:/temp/freetype/builds/win32/vc2010/freetype.vcxproj /p:PlatformToolset=v100 /p:Configuration="Release Multithreaded" /p:PlatformTarget=x86 /p:ConfigurationType=StaticLibrary /p:TargetName=freetype /p:OutDir="C:/temp/freetype/lib32/" /p:IntDir="C:/temp/freetype/obj32/" >> freetype.log

3. Ensure the parameters for x86/32bit is correctly configured as follows:

bash configure --disable-ccache
--with-boot-jdk=/cygdrive/c/temp/jdk7
--with-freemarker-jar=/cygdrive/c/temp/freemarker.jar
--with-freetype-include=/cygdrive/c/temp/freetype/include
--with-freetype-lib=/cygdrive/c/temp/freetype/lib32
--target=x86 --with-target-bits=32

[sxa]

build-softlayer-win2012r2-x64-1 has been tagged with x86-32 to allow me to prototype changes.

In addition to the freetype 32-bit compilation (I've used 2.5.3 as that was the one already on the machine)

I've also had to execute:

regsvr32 "C:\Program Files (x86)\Microsoft Visual Studio 10.0\DIA SDK\bin\msdia100.dll"

[ChengJin01]

Given that all issues with freeType were fixed, the remaining configuration should be fine with --target=x86 --with-target-bits=32 before compilation (we already made changes at Eclipse/OpenJ9 to get this mk/config scripts work).

[sxa]

Initial run now executing at https://ci.adoptopenjdk.net/view/work%20in%20progress/job/openjdk8_openj9_build_x86-32_windows

[gdams]

FYI you'll need to modify the release job here https://github.com/AdoptOpenJDK/openjdk-website-backend/blob/master/sbin/Release.sh#L48

[ChengJin01]

.../OpenJDK8_x64_Windows_201824051125/jdk8u172-b11
$ jre/bin/java -version
openjdk version "1.8.0_172-adoptopenjdk"
OpenJDK Runtime Environment (build 1.8.0_172-adoptopenjdk-_2018_05_24_11_28-b00)
Eclipse OpenJ9 VM (build master-44d1a2e9, JRE 1.8.0 Windows 7 x86-32-Bit 20180524_4 (JIT enabled, AOT enabled)
OpenJ9 - 44d1a2e9
OMR - a8771355
JCL - 556b4d29 based on jdk8u172-b11)

It looks good from the version output. Just the package name seems incorrect: "OpenJDK8_x64_Windows_201824051125".
it should be e.g. OpenJDK8_x86_Windows_201824051125 or OpenJDK8_x86-32_Windows_201824051125

[sxa]

Initial builds created at https://ci.adoptopenjdk.net/view/work%20in%20progress/job/openjdk8_openj9_build_x86-32_windows/

(Also hotspot job created but not yet succeeding at https://ci.adoptopenjdk.net/view/work%20in%20progress/job/openjdk8_build_x86-32_windows/)

[ChengJin01]

_sanity.functional_jdk8u172-b11_openjdk8_openj9_build_x86-32_windows.txt

I ran sanity tests from functional/Java8andUp locally with the build created at https://ci.adoptopenjdk.net/view/work%20in%20progress/job/openjdk8_openj9_build_x86-32_windows/ (via Cygwin on a Win 7 Pro 64-bit OS) and all tested buckets passed except one which doesn't exist as follows:

TOTAL: 38 EXECUTED: 28 PASSED: 27 FAILED: 1 SKIPPED: 10
(the only failing buckets is AttachAPISanity_SE80_0 which doesn't exist)

So, it seems the build functionally works good as expected.

[sxa]

I have also verified that the builds work on a purely 32-bit OS (Well, they get a far as running java -version correctly :-)

[ChengJin01]

@smlambert @llxia , is there any other test we are able to run to verify the 32-bit build created at https://ci.adoptopenjdk.net/view/work%20in%20progress/job/openjdk8_openj9_build_x86-32_windows/ (not manually run) ?

[llxia]

No, I think that is the only one we have so far. @sophia-guo is working on wins builds. She maybe able to tweak something for you, but we do not have one ready to use. As far as I know, the win machine configuration is not finalized. AdoptOpenJDK and OpenJ9 wins machines are configured differently.

[ChengJin01]

https://ci.adoptopenjdk.net/view/all/job/Grinder/39/consoleFull
regression_sanity_tests.txt

20:36:08 Test results: passed: 308; failed: 14
20:41:30 Test results: passed: 417; failed: 13
20:43:00 Test results: passed: 34
20:46:55 Test results: passed: 382; failed: 29; error: 1
20:54:14 Test results: passed: 315; failed: 10
21:07:22 Test results: passed: 574; failed: 27
21:17:11 Test results: passed: 123; failed: 10

21:17:33 +++++++++++++++++++++++++++
21:17:33 PASSED test targets:
21:17:33 	jdk_math_0
21:17:33 
21:17:33 FAILED test targets:
21:17:33 	jdk_io_0
21:17:33 	jdk_lang_0
21:17:33 	jdk_net_0
21:17:33 	jdk_nio_0
21:17:33 	jdk_util_0
21:17:33 	jdk_rmi_0
21:17:33 
21:17:33 TOTAL: 7   EXECUTED: 7   PASSED: 1   FAILED: 6   SKIPPED: 0

@sophia-guo already helped to run the regression (sanity) tests at ci.adoptopenjdk.net with the created build above. It seems most of failing teat cases required Cygwin be there to get executed (currently not run via Cygwin & no support on Cygwin temporarily).

The remaing failing cases were io/http/rmi related:

20:35:31 TEST RESULT: Failed. Execution failed: `main' threw exception: java.lang.Exception: Failed test: Should skip 164, really skipped 160lines (openjdk_regression\openjdk-jdk\jdk\test\java\io)
20:43:47 TEST RESULT: Failed. Execution failed: `main' threw exception: java.lang.RuntimeException: wrong amount of data returned (openjdk_regression\work\classes\X\com\sun\net\httpserver)
20:43:50 TEST RESULT: Failed. Execution failed: `main' threw exception: java.lang.RuntimeException: wrong amount of data returned (openjdk_regression\work\classes\X\com\sun\net\httpserver)
20:44:30 TEST RESULT: Failed. Execution failed: `main' threw exception: java.lang.RuntimeException: error (openjdk_regression\work\classes\X\com\sun\net\httpserver)
20:44:35 TEST RESULT: Failed. Execution failed: `main' threw exception: java.lang.RuntimeException: error (openjdk_regression\work\classes\X\com\sun\net\httpserver)
20:45:25 TEST RESULT: Failed. Execution failed: `main' threw exception: java.lang.RuntimeException: test failed (openjdk_regression\openjdk-jdk\jdk\test\java\net)
21:11:52 TEST RESULT: Failed. Execution failed: `main' threw exception: java.lang.RuntimeException: TEST FAILED: export on already-bound port succeeded (openjdk_regression\work\classes\2\java\rmi\server\RemoteServer)

[ChengJin01]

Already confirmed with @smlambert, part of openjdk_regression, system and functional tests will be executed on the created build on a weekly basis and might go daily, depending on how much spare windows machines are available for tests.

[ChengJin01]

The Win 32-bit has been added to the openjdk8_openj9 nightly build pipelines via #355

[smlambert]

Yes, though based on the build failure mentioned above, tonights run will fail, and not trigger any testing. Once the build issue is corrected and build jobs complete successfully, a subset of ~2900 openjdk regression tests will be triggered against them (we can triage those, and later decide if we should add system tests to the set or not).

[smlambert]

openjdk8-openj9 Win32 nightly builds are now being published to the website, this can now be closed.