JDK-8253947: JEP 388 Windows/AArch64 backport to jdk11u

[Bernhard Urban-Forster]

Thanks Vladimir and Andrew!

> 1) support for macos/aarch64 is not just jep-391, we split our jep into pieces,
> some of them were macos/aarch64 independent and were pushed to upstream 
> as separate commits ( for example, macos MAP_JIT support). also some
> important pieces from community ( like JNF dependency removal by Phil Race).
> All these pieces need to be backported before jep-391.

For JEP 388 we have a similar situation, although those independent changes are
at a smaller scale I would say. For example let's have a look at JDK-8248403 [1],
does it need its own PR on the jdk11u repository or is it fine package it all
together in one big PR?

Here is a rebased jdk11u branch btw:
https://github.com/lewurm/jdk11u/commits/jdk11u-windows


> 2) we have a plan to go in a traditional way, ojdk17 -> ojdk15 -> ojdk13
> -> ojdk11

I would say jdk13u/jdk15u is low-priority for us, but I can have a look at them
for JEP 388 once I've figured out the process for jdk11u (assuming it's the same).


> Cross compilation is not a requirement for a backport.

Cross compilation is the only way to build Windows+Arm64 today, so we need
at least the hacky variant that I suggested.


Thanks,
-Bernhard

[1] https://bugs.openjdk.java.net/browse/JDK-8248403

________________________________________
From: Vladimir Kempik
Sent: Wednesday, March 24, 2021 09:53
To: Bernhard Urban-Forster
Cc: Andrew Haley; Anton Kozlov; aarch64-port-dev at openjdk.java.net; Monica Beckwith; jdk-updates-dev at openjdk.java.net; Magnus Ihse Bursie; openjdk-aarch64
Subject: Re: JDK-8253947: JEP 388 Windows/AArch64 backport to jdk11u


Hello
there are few caveats with backporting jep-391:

1) support for macos/aarch64 is not just jep-391, we split our jep into pieces, some of them were macos/aarch64 independent and were pushed to upstream as separate commits ( for example, macos MAP_JIT support). also some important pieces from community ( like JNF dependency removal by Phil Race). All these pieces need to be backported before jep-391.

2) we have a plan to go in a traditional way, ojdk17 -> ojdk15 -> ojdk13 -> ojdk11

If you don’t want to bother with jdk15/13 I would suggest you to go without waiting for jep-391.

Regards , Vladimir

> 24 марта 2021 г., в 00:35, Bernhard Urban-Forster <beurba at microsoft.com> написал(а):
> 
> Hello,
> 
> Spinning off the discussion from here: https://github.com/openjdk/jdk/pull/2200#issuecomment-804927150
> 
> And further context, there have been some discussions about it before:
> https://mail.openjdk.java.net/pipermail/aarch64-port-dev/2020-October/009727.html
> 
> 
> Open questions:
> 
> 1. JEP 391 (macOS/AArch64) has a dependency on JEP 388, and I'm assuming we want both to be backported.  Question: Is it preferred to do it one go (both together), or should we do it separately (that would be Windows first, then macOS)?
> 
> 2. JEP 388 includes build changes to add cross compilation support for Windows in a hacky way. It's enough to get a build out of it [1], but it's not exactly clean. It was eventually cleaned up with the "WINENV" patch by Magnus [2], but imho it isn't trivial to backport that change.  Thoughts?
> 
> 
> Assuming the answer to question 1 is to split the backports: I think https://github.com/openjdk/aarch64-port/tree/jdk11-windows by Ludovic is in an okay shape, but I will take it for a spin tomorrow and then submit an initial PR against https://github.com/openjdk/jdk11u
> 
> 
> Thank you,
> -Bernhard
> 
> 
> [1] with workarounds: https://github.com/openjdk/jdk/pull/212#issuecomment-695024586
> [2] https://github.com/openjdk/jdk/pull/1597