RFR(S): 8022177: Windows/MSYS builds broken

Erik Joelsson erik.joelsson at oracle.com
Wed Aug 6 10:39:32 UTC 2014

I discovered another problem for msys use in 
make/common/NativeCompilation.gmk. The logic that rewrites the output 
from "cl.exe -showIncludes" into makefile dependencies assumes cygwin. 
This will need fixing to support incremental builds on msys.


On 2014-08-06 11:36, Erik Joelsson wrote:
> I gave msys another shot and managed to get it running on my laptop, 
> even with mintty so it's not driving me nuts trying to interact with 
> it. With your patch I managed to get through configure. When building, 
> it seems make just stops after a bit in langtools however, hogging one 
> full cpu core, at least at default concurrency setting (which is 4 on 
> my machine). Running with JOBS=1 seems to work better.
> Are you having the same problems, only being able to build without 
> concurrency?
> /Erik
> On 2014-08-06 11:01, Erik Joelsson wrote:
>> Hello Volker,
>> The changes seem reasonable to me, consider them reviewed. I can make 
>> the push if you like.
>> On msys, I would really like to be able to test this myself, but I 
>> find it hard to get a proper environment setup. Can you point me to a 
>> good guide on what I need to do?
>> /Erik
>> On 2014-08-05 20:26, Volker Simonis wrote:
>>> Hi,
>>> could somebody please review these straightforward changes which fix
>>> the MinGW7MSYS build on Windows:
>>> http://cr.openjdk.java.net/~simonis/webrevs/8022177/
>>> https://bugs.openjdk.java.net/browse/JDK-8022177
>>> The details of the change can be found below and in the webrev. I
>>> think I need a sponsor for this change because it requires the
>>> re-generation of generated-configure files (closed and open).
>>> I also want to mention that we at SAP are using the MinGW/MSYS build
>>> on Windows since more than a year for all our Java builds (from 4 to
>>> 8) on Windows and we're highly committed in keeping this build
>>> environment alive.
>>> Independently from this change I also have a general question related
>>> to the MSYS build:
>>> During an MSYS build the Windows command shell is usually called 
>>> like this:
>>> `cmd //c echo $unix_path`
>>> As far as I understand, the double slash is needed to prevent MSYS
>>> from rewriting '/c' to 'c:/' when calling a non-MSYS executable. But
>>> sometimes (notably in BASIC_MAKE_WINDOWS_SPACE_SAFE_MSYS in
>>> basic_windows.m4) we call the Windows command shell without a quoted
>>> forward slash like this:
>>> new_path=`cmd /c "for %A in (\"$input_path\") do @echo %~sA"
>>> and I don't understand why this works.
>>> To complicate things, the MSYS environment contains its own 'cmd'
>>> shell script in '/usr/bin/cmd' which only redirects to the
>>> corresponding Windows command shell like so:
>>> #!/bin/sh
>>> "$COMSPEC" "$@"
>>> Unfortunately I didn't succeed in unifying all the calls to 'cmd' to a
>>> single pattern (i.e. either `cmd //c ...`, or `/usr/bin/cmd /c ...`)
>>> which would be desirable, because depending on the actual PATH
>>> settings, `cmd` may either call "/usr/bin/cmd" or
>>> "C:\Windows\System32\cmd.exe".
>>> Can anybody shed some light on this topic (i.e. which "cmd" we want to
>>> call and why we have to quote "/c" differently)?
>>> Thank you and best regards,
>>> Volker
>>> 1. common/autoconf/basics.m4
>>> - Check for bsdcpio if cpio is not found. Also make the check for cpio
>>> optional, as suggested in the bug report.
>>> - Only do the BASIC_CHECK_SRC_PERMS check on Cygwin as it only seems
>>> to be relavant there and because the stat utility isn't availabel in
>>> MSYS anyway.
>>> 2. common/autoconf/basics_windows.m4
>>> - Same story like on Cygwin: '.bat' and '.cmd' files are not always
>>> considered executable in MSYS causing which to not find them.
>>> 3. common/autoconf/flags.m4
>>> - Use "-WX" option syntax for cl options (instead of "/WX") to prevent
>>> option rewriting by MSYS.
>>> 4. toolchain_windows.m4
>>> - Account for the fact that the MSYS file utility reports different
>>> file types for Windows DLLs than its Cygwin counterpart.

More information about the build-dev mailing list