Round two - RFR (7152336): Enable OpenJDK builds on Windows with MinGW/MSYS

[Volker Simonis]

Hi Tim,

I'm still glad that you're pushing this forward:)

Just wanted to let you know two more things we found out about the
MinGW build in the hope they may be useful for others as well:

 - you can considerably speed up  the build (especially an incremental
build which only rebuilds a few files) if you disable implicit make
rules (i.e. use make's '-r' option). You can set "MAKE=make -r" to
also use this option for recursive make invocations but perhaps it
would be better to just define "MAKEFLAGS += -r" in the top-level
makefiles.

Without this option, make tries to look up a lot of non-existing RCS
and SCCS files which especially hurts on Windows (see
http://www.electric-cloud.com/blog/2009/08/19/makefile-performance-built-in-rules/
for a more detailed description).

- internally we use a sh-script (build.sh) which sets some environment
variables and then calls make to start the build. Under MKS it was
possible to simply call such a sh-script from an arbitrary DOS-shell
because MKS registered '.sh' as a known extension and just did the
right thing.

Under MinGW/MSYS we had to explicitly call 'bash.exe -i build.sh
<shell-script-args>'. However this unfortunately didn't work reliable.
Instead the top-level DOS-shell crashed (with the error "Command
Windows Host has stopped working") during the build which results in
thousands of DOS-boxes popping up for every executed system command.
This problem can be solved by just piping the desired command to the
shell like so: "build.sh <shell-script-args> | bash.exe -i"

Regards,
Volker

On Wed, Jun 27, 2012 at 6:45 PM, Tim Bell <tim.bell at oracle.com> wrote:
> I went over the open changes again and removed some debug code left in
> by mistake, updated copyright dates and fixed a few typos.  Here is an
> updated webrev:
>
>    http://cr.openjdk.java.net/~tbell/7152336/webrev.01/
>
> These changes are building in both 32 and 64 bit mode on my Windows 7
> system.
>
> jtreg test runs of the '-automatic -noshell' tests under test/java/lang
> test/java/math test/java/util ran as expected (2 failures, both known bugs).
>
> Thanks in advance for your review and feedback - I'd like to get these
> changes in soon.
>
> Tim Bell
>
>
> On 06/13/12 22:31, Tim Bell wrote:
>>
>> Hello everyone-
>>
>> Kelly asked me to pick up on bug #/7152336 "//Enable builds on Windows
>> with MinGW/MSYS"/, and this email thread:
>>
>>
>> http://mail.openjdk.java.net/pipermail/build-dev/2012-April/thread.html#6083
>>
>> As David pointed out, we will need at least one other bug # for the
>> hotspot changes.  That said, this is enough to get me started.
>>
>> Hi Volker:
>>
>> I have applied the patches originally from your posting.  Many thanks for
>> that:
>> http://cr.openjdk.java.net/~simonis/MinGW_MSYS.v1/
>>
>> With a few modifications (keep cpio for non MinGW/Msys builds, keep MKS as
>> an option), the proposed changes are visible here for review:
>>
>>  http://cr.openjdk.java.net/~tbell/7152336/webrev.00/
>>
>> For reference, my test build log is visible here:
>>
>>
>> http://cr.openjdk.java.net/~tbell/7152336/webrev.00/full_control_build_no_docs.log
>>
>> Additional test builds on JPRT (our internal build apparatus) verified
>> that I didn't regress the existing build.
>>
>> Abbreviated jtreg [1] testing on this build was successful:
>>
>> $ /d/tools/jdk8/7152336/windows-i586/bin/java -jar
>> /d/tools/jtreg-internal/jtreg/lib/jtreg.jar -automatic -noshell
>> test/java/lang test/java/math test/java/util
>> Directory "JTreport" not found: creating
>> Directory "JTwork" not found: creating
>> Directory "JTwork\scratch" not found: creating
>> Test results: passed: 698; failed: 1; error: 5
>> Report written to D:\tools\jdk8\7152336\jdk\JTreport\html\report.html
>> Results written to D:\tools\jdk8\7152336\jdk\JTwork
>> Error: Some tests failed or other problems occurred.
>>
>> The failing test (java/lang/Math/WorstCaseTests.java) is due to a known
>> regression: 7174532
>> <http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7174532>
>> "jdk/test/java/lang/Math/WorstCaseTests.java failing on x86"
>>
>> The 5 error tests are all ignored until bug xxxxxxx (for some value of x)
>> is resolved.
>>
>> Thanks in advance for your review and feedback -
>>
>> Tim Bell
>>
>> [1]  http://openjdk.java.net/projects/code-tools/
>>
>