[Performance] Reduce allocations in TextShaperImpl by reusing HarfBuzz Buffer

[dme-compunet]

What does the pull request do?

Currently, a new HarfBuzz Buffer is allocated in TextShaperImpl for every text shaping operation. This can be avoided by reusing a single buffer and calling Reset before each shaping operation.

What is the current behavior?

What is the updated/expected behavior with this PR?

How was the solution implemented (if it's not obvious)?

Checklist

• Added unit tests (if possible)?
• Added XML documentation to any related classes?
• Consider submitting a PR to https://github.com/AvaloniaUI/avalonia-docs with user documentation

Breaking changes

Obsoletions / Deprecations

Fixed issues

[Gillibald]

Have you measured this? Resetting the buffer might be more costly.

[Gillibald]

It looks like buffer create calls reset, so this should be fine: https://github.com/harfbuzz/harfbuzz/blob/main/src/hb-buffer.cc#L762

[Gillibald]

LGTM

[avaloniaui-bot]

You can test this PR using the following package version. 12.0.999-cibuild0056607-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

[kerams]

Is there a chance that this runs on ephemeral threads, and leaks memory over time since there is no dispose? Or does the runtime automatically dispose of thread static values, if possible, when a thread is destroyed?

[Gillibald]

We can make the ITextShaper disposable and the buffer does not need to be static.

[MrJul]

Is there a chance that this runs on ephemeral threads, and leaks memory over time since there is no dispose? Or does the runtime automatically dispose of thread static values, if possible, when a thread is destroyed?

The runtime won't dispose of the values, but will finalize them, which would reclaim the native resources.

Anyways, I wouldn't worry too much: 99.9% of text shaping operations will occur on the UI thread. We're already caching some values using thread static fields in other parts of the text layout engine, for example in TextFormatterImpl.