[JSC] Do not use bytecode cache on $.agent worker threads

[gezalore]

20ac310

[JSC] Do not use bytecode cache on $.agent worker threads

Patch by Geza Lore <glore@igalia.com > on 2022-05-23
https://bugs.webkit.org/show_bug.cgi?id=240642

Reviewed by Yusuke Suzuki.

Workers started via $.agent.start are not shut down in a synchronous
manner, and it is possible the main thread terminates the process while
a worker is writing its bytecode cache, which results in intermittent
test failures. As $.agent.start is only a rarely used testing facility,
we simply do not cache bytecode on these threads.

Also un-skip test on ARMv7 that used to fail because of this.

* Source/JavaScriptCore/jsc.cpp:
(Worker::isMain const):
(Worker::Worker):
(runJSC):
 * JSTests/stress/lars-sab-workers.js:

Canonical link: https://commits.webkit.org/250858@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@294632 268f45cc-cd09-0410-ab3c-d52691b4dbfc

[Constellation]

r=me

[kmiller68]

Can we make it so other threads can read from the cache but not write? That way we can test the thread safety of using the caching system (at least partly).

[gezalore]

@kmiller68 The agent threads execute different code from the main thread, so whatever the main thread writes to the cache is not what the agent threads would try to read, so you would never have a cache hit, so I don't see much point int trying to do that.

[gezalore]

@kmiller68 What would you like me to do? Can we unblock this please?

Sounds like my chance won’t help. Unblocking.

[webkit-early-warning-system]

Committed r294632 (250858@main): https://commits.webkit.org/250858@main

Reviewed commits have been landed. Closing PR #776 and removing active labels.