RFR: 8250876: Build system preparation to macos on aarch64

Mon Aug 10 18:19:05 UTC 2020

Looks good.

/Erik

On 2020-08-10 07:14, Vladimir Kempik wrote:
> Hello
> Please check the updated webrev - http://cr.openjdk.java.net/~vkempik/8250876/webrev.02/
> Thanks, Vladimir
>
>> 10 авг. 2020 г., в 16:11, Erik Joelsson <erik.joelsson at oracle.com> написал(а):
>>
>> On 2020-08-10 03:47, Vladimir Kempik wrote:
>>> Hello
>>>
>>> Renamed the bug to "Fix issues with cross-compile on macos"
>>>
>>> Please check updated webrev: http://cr.openjdk.java.net/~vkempik/8250876/webrev.01/
>>>
>>> Adlc is fine with cross-compiling now.
>>>
>> While not strictly needed for ADLC, I would like to see BUILD_CC getting the same treatment here for consistency.
>>
>> /Erik
>>
>>> Regards, Vladimir
>>>> 10 авг. 2020 г., в 12:52, Magnus Ihse Bursie <magnus.ihse.bursie at oracle.com> написал(а):
>>>>
>>>> So, basically, what this patch is about is not so much "preparation for aarch64" as "allow cross-compile on macos"?  If I understand you correctly, maybe you should rename the bug?
>>>>
>>>> /Magnus
>>>>
>>>> On 2020-08-04 16:09, Erik Joelsson wrote:
>>>>> That's a good find! You are correct in that we haven't cross compiled in any direction involving Macosx before.
>>>>>
>>>>> The preferred patch would be a bit more elaborate than that. Ideally we need better control over the toolchain type of the BUILD_* compiler settings. For now, I think just forcing clang/clang++ if BUILD_OS is macosx is good enough.
>>>>>
>>>>> /Erik
>>>>>
>>>>> On 2020-08-04 07:02, Bernhard Urban-Forster wrote:
>>>>>> Good observation David, the change in adlc is just fixing a symptom. The difference to a regular macOS build is that technically, despite running on the same machine, it's actually cross compiling due to Rosetta being the --build=x86_64 system.
>>>>>>
>>>>>> Being a cross-compile, we therefore hit this case:
>>>>>> https://github.com/openjdk/jdk/blob/b0ceab23dd4176329cbf3a95f21e8e9ac2d8723f/make/autoconf/toolchain.m4#L905-L921
>>>>>>
>>>>>> And thus infers `/usr/bin/CC` for CXX.
>>>>>>
>>>>>> I guess cross compiling hasn't been a thing on macOS yet. I tried the following and it passes the adlc build:
>>>>>>
>>>>>> --- a/make/autoconf/toolchain.m4
>>>>>> +++ b/make/autoconf/toolchain.m4
>>>>>> @@ -917,7 +917,7 @@ AC_DEFUN_ONCE([TOOLCHAIN_SETUP_BUILD_COMPILERS],
>>>>>>        # find the build compilers in the tools dir, if needed.
>>>>>>        UTIL_REQUIRE_PROGS(BUILD_CC, [cl cc gcc])
>>>>>>        UTIL_FIXUP_EXECUTABLE(BUILD_CC)
>>>>>> -    UTIL_REQUIRE_PROGS(BUILD_CXX, [cl CC g++])
>>>>>> +    UTIL_REQUIRE_PROGS(BUILD_CXX, [clang++ cl CC g++])
>>>>>>        UTIL_FIXUP_EXECUTABLE(BUILD_CXX)
>>>>>>        UTIL_PATH_PROGS(BUILD_NM, nm gcc-nm)
>>>>>>        UTIL_FIXUP_EXECUTABLE(BUILD_NM)
>>>>>>
>>>>>> Although I'm not sure about its cleanliness :-)
>>>>>>
>>>>>> -Bernhard
>>>>>>
>>>>>> ________________________________________
>>>>>> From: build-dev <build-dev-retn at openjdk.java.net> on behalf of David Holmes <david.holmes at oracle.com>
>>>>>> Sent: Tuesday, August 4, 2020 00:46
>>>>>> To: Erik Joelsson; Vladimir Kempik; build-dev
>>>>>> Cc: Anton Kozlov; Alexander Ioffe; Andrew Brygin; Andrey Petushkov
>>>>>> Subject: Re: RFR: 8250876: Build system preparation to macos on aarch64
>>>>>>
>>>>>> On 3/08/2020 10:57 pm, Erik Joelsson wrote:
>>>>>>> Hello Vladimir,
>>>>>>>
>>>>>>> These changes look innocent enough to me. They aren't actually adding
>>>>>>> macosx-aarch64 support, they are just removing two minor (and more
>>>>>>> likely OS version related) hurdles from the build. You still have to
>>>>>>> provide the actual configuration on the configure command line as is
>>>>>>> shown in your example. Before we can call build system support, we would
>>>>>>> need configure to automatically setup those flags and add a separate
>>>>>>> parameter for the JNF framework. So, given that, I don't think this
>>>>>>> change warrants a JEP in itself.
>>>>>> Of course this change doesn't warrant a JEP in itself :) My point is
>>>>>> that until we have a JEP for the platform that is being targeted then we
>>>>>> shouldn't be making changes to provide support for that platform.
>>>>>>
>>>>>> That said I didn't actually look at the changes but focused on the
>>>>>> larger stated aim, so apologies for that.
>>>>>>
>>>>>> The actual changes here are small and not obviously related to
>>>>>> supporting macOS-Aarch64. But I'm unclear on this change as it affects
>>>>>> all macOS builds:
>>>>>>
>>>>>>      42   else ifeq ($(call isBuildOs, macosx), true)
>>>>>>      43     ADLC_LDFLAGS := -lc++
>>>>>>
>>>>>> if this was fixing a bug as indicated, why do we not see this bug in
>>>>>> regular builds?
>>>>>>
>>>>>> Thanks,
>>>>>> David
>>>>>> -----
>>>>>>
>>>>>>
>>>>>>> My only complaint is that you revert jib-profiles.js. That file is only
>>>>>>> used internally at Oracle. If/when we need it to support macosx-aarch64,
>>>>>>> we will provide those changes.
>>>>>>>
>>>>>>> I must say I'm happy to see you managed to get a working build
>>>>>>> configuration with just this though!
>>>>>>>
>>>>>>> /Erik
>>>>>>>
>>>>>>> On 2020-08-01 00:24, Vladimir Kempik wrote:
>>>>>>>> Hello
>>>>>>>>
>>>>>>>> Please review this change for JDK-8250876
>>>>>>>>
>>>>>>>> This changeset adds support for macos/aarch64 into build system.
>>>>>>>> It will allow to crosscompile for macos/aarch64 using intel mac as well.
>>>>>>>>
>>>>>>>> This changeset does NOT address some arm specific issues in the macos
>>>>>>>> related code, we plan to do that in s separate commit.
>>>>>>>>
>>>>>>>> An example of configure to cross-compile for macos/arm64:
>>>>>>>>
>>>>>>>> --with-boot-jdk=/path/to/java/
>>>>>>>> --with-build-jdk=/path/to/same/java/as/compiled
>>>>>>>> --disable-warnings-as-errors --with-jvm-variants=zero
>>>>>>>> --openjdk-target=aarch64-apple-darwin --with-extra-cflags='-arch
>>>>>>>> arm64' --with-extra-ldflags='-arch arm64
>>>>>>>> -F/Path/To/Folder/Containing/JNF_framework/'
>>>>>>>> —with-extra-cxxflags='-arch arm64’
>>>>>>>>
>>>>>>>> JNF.framework is missing arm64 part as of next macos release, but
>>>>>>>> Apple has opensourced it.
>>>>>>>>
>>>>>>>> Fix to adlc were needed due to it using symbols from stdc++ and not
>>>>>>>> linking to it, so it fails when doing make images.
>>>>>>>>
>>>>>>>> The webrev: https://nam06.safelinks.protection.outlook.com/?url=http:%2F%2Fcr.openjdk.java.net%2F~vkempik%2F8250876%2Fwebrev.00%2F&data=02%7C01%7Cbeurba%40microsoft.com%7C0c8d58d5eb9144e8717f08d837ff3736%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637320916565796801&sdata=HpXJmHXbuawTdExWESK9ssesYTuPTj7N6inXjaHfVaM%3D&reserved=0
>>>>>>>> The bug: https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fbugs.openjdk.java.net%2Fbrowse%2FJDK-8250876&data=02%7C01%7Cbeurba%40microsoft.com%7C0c8d58d5eb9144e8717f08d837ff3736%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637320916565796801&sdata=9z2Nw8d0pa5huxUKOYorMOVy6SBo7o%2FhDT1EmgOhxQ8%3D&reserved=0
>>>>>>>>
>>>>>>>> Testing: jdk/submit.
>>>>>>>>
>>>>>>>> Thanks, Vladimir.