Fix two issues in --spill-pointers

[juj]

Fix two issues in --spill-pointers: 1) properly generate the stack frame in a stack-grows-downwards manner, and 2) do not spill pointers in the functions stackAlloc()/stackSave()/stackRestore(), which should operate on the global stack pointer without a stack frame of their own.

[hoodmane]

What is the status of this patch? Has anyone investigated the CI failures? Thanks for working on this @juj! It looks helpful.

[juj]

Took quite a bit of time to get back to looking into GC on Emscripten..

I see that in current main branch, the issue 1) has already been fixed.

So I reset this branch now on top of latest main, and only kept the second fix

"2) do not spill pointers in the functions stackAlloc()/stackSave()/stackRestore(), which should operate on the global stack pointer without a stack frame of their own."

from the original code.

I am developing a small toy garbage collector to test things out, and with this Binaryen PR landed, it is now working properly to scan the local stack and global static data section.

[juj]

I am testing this out against https://github.com/juj/emgc to prototype garbage collection against Emscripten. The plan is to experiment with multithreaded garbage collection from multiple managed threads, with cooperative stop-the-world signaling, and then try to transplant the whole thing back to Boehm, or something like that.

[kripken]

@juj Are you aiming to get this approved and landed here in upstream, or is this just for experimentation at this time?

[juj]

This is now for experimentation, though I'll try to work towards test cases on the stackAlloc issue to make sure I'm proposing a fix that patches that issue.