Safer multisample detection

[dev7355608]

Description of change

Should fix #9269.

Number of samples is chosen based on the internal format. Samples above MAX_SAMPLES are filtered out as well as samples that are not valid depth/stencil samples.

Changed the type of GLFramebuffer.multisample to number, because technically it isn't necessarily a value in MSAA_QUALITY.

Pre-Merge Checklist

• Tests and/or benchmarks are included
• Documentation is changed or added
• Lint process passed (npm run lint)
• Tests passed (npm run test)

[codesandbox-ci]

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

Latest deployment of this branch, based on commit 1008800:

Sandbox	Source
pixi.js-sandbox	Configuration

[dev7355608]

Before I mark this ready-for-review: Should we change how detectSamples works? If the valid samples are [8, 4] then with MSAA_QUALITY.LOW there's no multisampling. Maybe we should select the minimum of the samples that are greater or equal to the number of requested samples. Then LOW would map to 4 in the the case of [8, 4]. That might make more sense considering that the samples parameter of renderbufferStorageMultisample is just a minimum: the implementation is free choose a higher number of samples than requested.

[dev7355608]

Before I mark this ready-for-review: Should we change how detectSamples works? If the valid samples are [8, 4] then with MSAA_QUALITY.LOW there's no multisampling. Maybe we should select the minimum of the samples that are greater or equal to the number of requested samples. Then LOW would map to 4 in the the case of [8, 4]. That might make more sense considering that the samples parameter of renderbufferStorageMultisample is just a minimum: the implementation is free choose a higher number of samples than requested.

I'll leave this for another PR.

[dev7355608]

I put the Framebuffer.multisample changes in a separate PR (#9390) wondering why I didn't in the first place, but now I recall: without the changes here #9390 does of course nothing, because the samples are detected and set in initFramebuffer. With the changes here the samples are detected in updateFramebuffer, and so it makes sense for Framebuffer.multisample to increase dirtyId.

[GoodBoyDigital]

sorry it took so long to get round to this! all makes sense! one tiny comment. Could we move the logic out to a getMaxSamples() function? - but other than that good to go!