[3.x] SurfaceTool - efficiency improvements

[lawnjelly]

Changed to use LocalVector rather than linked lists.

This is actually part of #61568 but has been separated for easier review.

Notes

• This is much faster than the previous code (and speed is critical for merging).
• Conversion of SurfaceTool to use vectors rather than linked lists has already been done in master.

Areas of interest

• Vertex now has fixed arrays for bones and weights, this allows more efficient access / movement within LocalVectors, and we don't support more than 4 bones in 3.x anyway.
• Smoothing groups implementation is slightly altered but the end result should be the same.

[pseidemann]

hi,
nice work!

just wanted to tell you there is an open pr #68034 (for 4.0) which changes the implementation of the smooth groups. not sure if it's relevant here

[lawnjelly]

just wanted to tell you there is an open pr #68034 (for 4.0) which changes the implementation of the smooth groups. not sure if it's relevant here

Although the PR you linked looked sensible, I'm not attempting to make any changes to smoothing behaviour in this PR. Any changes should be in a separate PR considered on their own merit. This makes things easier to review. 👍

[clayjohn]

hi, nice work!

just wanted to tell you there is an open pr #68034 (for 4.0) which changes the implementation of the smooth groups. not sure if it's relevant here

In addition to lawnjelly's comment, note that this PR is targetting 3.x, so that PR is totally irrelevant as 3.x and 4.0 work differently with respect to smooth groups

[pseidemann]

my small hope was that maybe it's possible to consolidate some backporting work with this or to stay more compatible to 4.0 (or the other way around) to make it easier to backport things later

[lawnjelly]

my small hope was that maybe it's possible to consolidate some backporting work with this or to stay more compatible to 4.0 (or the other way around) to make it easier to backport things later

Any changes to 3.x has to be backward compatible, so we can't easily change the existing behaviour. Nevertheless if there was a largely improved smoothing group implementation we could probably add it in parallel (as e.g. a new method) if there was demand and it had proved successful in 4.x.

[clayjohn]

Overall the change looks fine to me. I left a few style nitpicks, but the implementation looks good.

[fire]

@clayjohn @lawnjelly Is this pr ready to go?

[lawnjelly]

@clayjohn @lawnjelly Is this pr ready to go?

Yup, changes made. 👍

[clayjohn]

Seems fine to me. I tested locally and couldn't detect any regressions.

I tested on a mesh made of 100,000 primitive points and this PR improved the speed of creating and committing the mesh noticeably (~15% based on rough measurements)

[akien-mga]

Thanks!