Fix memory leak when removing map

[kamil-sienkiewicz-asi]

Hey, so I digged a little in code to resolve this issue #9126 and it kinda seems like I found what the issue is (for a bare map, nothing was touched after mounting).

Tested it using FinalizationRegistry and heap snapshots

Before:
When Map was added and removed using map.remove() it was piling up in memory heap snapshot.

After my changes:
When Map is removed and another instance of map is being mounted, previous instance is garbage collected.

Which wasn't exactly what I wanted.
I suspect that another issue lies in:

src/ui/map.js

 3881  const gl = this.painter.context.gl;

in _authenticate -> getMapSessionAPI -> error callback
which holds this.painter forever as
const mapSessionAPI_ = new MapSessionAPI(); is created in global scope and never freed.

but that code is just below warning about terms of service, so I'm not touching that :p

For future issues: Do I have permission to modify/debug this code with good intentions? All changes that improve mapbox would be put out in public PRs to approve by mapbox team.

@mourner @stepankuzmin let me know if that works and if you could handle issue in code that is under TOS 🤝

[CLAassistant]

All committers have signed the CLA.

[kamil-sienkiewicz-asi]

@stepankuzmin any followup on that/is something blocking this issue to be resolved? test-render-mac-safari-dev seems like something random broke, can't re-run 🤔

[mourner]

Thank you, nice catch!

[mourner]

and if you could handle issue in code that is under TOS 🤝

While we're still looking into this (most likely, it's fine for external contributors to edit that in PRs too), let us add a fix while we confirm.

[stepankuzmin]

Sorry for the slow response and thanks for the fix @kamil-sienkiewicz-asi!

[kamil-sienkiewicz-asi]

@stepankuzmin Thanks for the merge, however I can still observe that map is not removed entirely. It's not piling up in memory, however const gl = this.painter.context.gl; is still holding previous map instance after removal. It's freed when new map is initialised.

If you could confirm that I can touch this code under TOS I can work on that in a meanwhile.

[stepankuzmin]

@kamil-sienkiewicz-asi Hmm, that's weird, I've checked with your example from #9126 (comment) and haven't seen dangling Map instances. But sure, please go ahead and we'll review your PR 👍

[kamil-sienkiewicz-asi]

@stepankuzmin #13116 as promised.