[webkit-dev] Proposal on retiring JIT on Windows

[Yusuke Suzuki]

The reason is that the patch is reverted because it broke Windows DFG. This is not acceptable to us since we have no EWS bots running tests.
Without active maintainers / EWS bots, we cannot land JIT changes since it can break Windows, and it becomes huge burden than before (previously there was a Windows EWS bot running tests).

This is unfortunate, but if nobody steps up as a maintainer of JSC JIT on Windows (it involves adding test-running EWS bots, bug fixes of JSC on Windows etc.), then I think only the solution is disabling this feature.

So, if we would like to keep it, we need someone who steps up as a maintainer of JIT on Windows, and setting up bots running tests on Windows on EWS.

-Yusuke

> On Mar 25, 2023, at 9:27 PM, Kirsling, Ross <Ross.Kirsling at sony.com> wrote:
> 
> This seems unfortunate and unexpected to me—I thought having a singular Windows port was supposed to mean reduced maintenance burden, since we don't have to divide the number of Windows maintainers into two.
> 
> Although reconciling FTL with Windows' calling convention is a challenge that no one's ever had the time to prioritize, we've had a fully working DFG on Windows for so many years. If we turn that off, I can't imagine it ever being revived again.
> 
> How come the discussion here is immediately about getting rid of such a large swath of functionality instead of starting with consideration of the EWS situation itself?
> 
> Ross
> From: Yusuke Suzuki via webkit-dev <webkit-dev at lists.webkit.org>
> Sent: Sunday, March 26, 2023 7:23:04 AM
> To: Fujii Hironori <fujii.hironori at gmail.com>; Brent Fulgham <bfulgham at apple.com>; Mark Lam <mark.lam at apple.com>
> Cc: WebKit Development <webkit-dev at lists.webkit.org>
> Subject: Re: [webkit-dev] Proposal on retiring JIT on Windows
>  
>> How about LLInt? LLInt has some Windows specific code.
>> Can I revert the change if the JSC team breaks Windows port even though we have no EWS nor maintainers?
> 
> I think, ultimately, we cannot guarantee that it is working given that there is no EWS bots running tests on Windows, it means it is not debuggable to us.
> Frequency of breakage on LLInt would be smaller than breakage on JIT. But if it happens, then I think reverting is not OK since nothing is doable to JSC team.
> 
> @Brent @Mark What is your thought and plan?
> 
> -Yusuke
> 
>> On Mar 25, 2023, at 3:02 PM, Fujii Hironori <fujii.hironori at gmail.com> wrote:
>> 
>> It sounds reasonable. I don't object to removing Windows JIT support.
>> 
>> How about LLInt? LLInt has some Windows specific code.
>> Can I revert the change if the JSC team breaks Windows port even though we have no EWS nor maintainers?
>> 
>> On Sun, Mar 26, 2023 at 6:52 AM Yusuke Suzuki via webkit-dev <webkit-dev at lists.webkit.org <mailto:webkit-dev at lists.webkit.org>> wrote:
>> Hello,
>> 
>> I would like to propose retiring JIT on Windows JavaScriptCore.
>> Recently, Apple Windows EWS bots get removed. As a result, there is no test-running EWS bots on Windows.
>> 
>> This can work for the other part of WebKit since other ports EWS bots are running tests. However this does not work well for JSC.
>> Compile-tests is not sufficient for JIT since JIT is dynamically generated. And JIT is very architecture and platform specific.
>> Windows has different ABI, different calling convention, and register usage. JIT on Windows has very specific things.
>> 
>> Now, because there is no running EWS bots on Windows, it is not possible to catch Windows specific JIT related issues before landing.
>> Recently, JSC DFG patch has been reverted because Windows gets broken[1]. But this puts high burden to JSC maintenance since
>> there is no way to test it before landing, and it makes DFG changes very hard due to Windows.
>> 
>> So, given that there is no active maintainers of Windows JSC JIT and no EWS bots running tests, I propose retiring JIT on Windows.
>> 
>> [1]: https://github.com/WebKit/WebKit/commit/58f0d9e4a395e0173e4d3f59888bf0e761cf6ce3
>> 
>> -Yusuke
>> _______________________________________________
>> webkit-dev mailing list
>> webkit-dev at lists.webkit.org <mailto:webkit-dev at lists.webkit.org>
>> https://lists.webkit.org/mailman/listinfo/webkit-dev
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-dev/attachments/20230325/2c015d77/attachment.htm>