RFC: minimal backport of 8182299 build on OSX 10 + Xcode 8

[Simon Tooke]

On 7/3/2019 4:03 PM, Andrew John Hughes wrote:
>
> On 03/07/2019 19:30, Ben Evans wrote:
>> Hi,
>>
>> As this appears to have been Warnock'd (
>> https://en.wikipedia.org/wiki/Warnock%27s_dilemma), allow me to give it
>> another kick.
>>
>> My personal take (for the little that it's worth) is: Given current
>> adoption rates and trends, JDK 8 is going to be with us for a long time.
>> Restricting the potential community size of developers working on OpenJDK 8
>> by:
>>
>> 1.) Requiring that people who develop on Macs jump through virtualisation
>> hoops and
>> 2.) Making the building of binaries a byzantine process requiring the
>> maintenance of obsolete Macs
>>
>> is bad, and we should stop doing it.
>>
>> Thanks,
>>
>> Ben
>>
>> On Fri, Jun 28, 2019 at 4:09 PM Simon Tooke <stooke at redhat.com> wrote:
>>
>>> Hello everyone,
>>>
>>> I would like to make a somewhat controversial proposal: to backport the
>>> minimal changes required to enable jdk8u to compile and run when built
>>> with the latest macOS developer tools.
>>>
>>> I realize this falls outside of aph's guidelines of "bug fixes only,
>>> (for now)" and in many ways is only a developer convenience, since I am
>>> not advocating (at this time) for this build to become the default
>>> supported for macOS [1].
>>>
>>> There changes do not affect the current mac build, which requires an old
>>> version of Xcode which doesn't run on modern releases of macOS, but they
>>> make it much easier for macOS hackers to work with jdk8.
>>>
>>> At this point, testing has been confined to bootstrapping the build with
>>> a jdk8 built using this patch, and to using this build to build a
>>> working Graal substrateVM.
>>>
>>> My version of the backport limits the scope of the 8182299 patches to
>>> the subset required to get the JDK up and running.  I don't propose
>>> backporting any changes to remove Clang warnings, etc.  Because of that,
>>> my changes are confined in scope.
>>>
>>> Potential long term benefits (if this build does seem healthy enough for
>>> production) are simplified macOS build platforms, a more modern compiler
>>> and perhaps higher performance.
>>>
>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8182656
>>>
>>> Webrev: http://cr.openjdk.java.net/~stooke/webrevs/xcodemacos.webrev/
>>>
>>> Thanks for your time,
>>>
>>> -Simon
>>>
>>> [1] first, potentially removes support for macOS 10.8, second, needs
>>> more testing.
>>>
>>>
>>>
> I saw it, but there's a CPU on right now. I'll have a look properly once
> that's out of the way.
Yes, I know this is very much on the back burner for now. 
> My initial thought is what is the subset you refer to? The problem with
> backporting bits of a change is that it then looks like that change is
> backported, but some parts of it are actually missing.

We could avoid referring to the initial patch altogether; a large part
of it (which I avoided) was simply getting rid of clang warning
messages.  I just went for the bits that stop the crash, and skipped the
rest.