Transition internals

[pushkine]

Following #4742
Unfortunately crossfading and the animate directive turned out to be incompatible with transitions by design. There is still a few things related to transitions from that PR, none of them are breaking

• New frame callback system keeps track of all pending tasks to shrink work performed each frame (previously iterating over a Set calling each animation) down to as little as one timestamp check

• Measurements and animations are now batched to be applied after flush, this avoids layout trashing and significantly improves performance in same tick tasks performing several DOM reads
  This is the cause of most changed tests as it was expected for animations to be applied synchronously
  Fixes Animation rects need to be measured before any other updates #4732

• Position freeze is now reversed if re-introed,
  fixes transition + animate where a node would hide behind others, fixes Elements (components) with transition overlap each other in an each block using flip animation #4910, fixes Animate breaks when used along with fly transition #6001

• Bidi transitions now recompute on every run, fixes Nested Transitions NOT Honored #4784, fixes Slide transition jumps on the way out if height changes while it's visible #4831

  • To further customization, if a bidi config returns a function it will now be called with a is_intro boolean, related Add direction parameter to transitions #4019

• Bidi transitions now handle several delay reversals, fixes delay in transitions only applies to last change #4968

  • Handling delay properly while also recomputing on every run requires to know whether a config has delay before running the transition function ( as its measurements cannot be delayed ), so on init and every time a config changes there is a dry run to scan for the delay parameter. If config returns a function it is called with is_intro undefined
    

• Fixed concurrency bugs related to bidi transitions reversal, fixes Error in bidi transitions when returning a function #4974, fixes outroing transition group never unmounts if one sibling remounts with delay in bidirectional transition #4954

• Added rotate option to fly transition, rewrote svelte/transition types

notes

• motion still uses the old loop system, I started rewriting some of it using motion stores from Transitions fix #4742 but I doubt anyone using the current implementation is concerned about performance
• transitions could be DRYer, and bidi are heavy on GC
• @rollup/plugin-typescript has to be updated to properly inline const enum

todo

Remove transition easing strategies ( cool but unnecessary )
Write tests for fixed bugs

Can't work on this for some time but it's there if someone needs it, it's all yours

[pushkine]

Also here's a playground for the fork https://svelte.dev/repl/3548e4d0a06f4440847e5c92caa02b93
Each column uses a different strategy ( the way easing is computed when transitions are reversed )
Changing easing does not work on the web repl since transitions are only ever computed once

always happy to instant chat on discord if you need me

[pushkine]

rereading it this looks like a memory leak, should push outroing.delete(block) into a transition group callback

[stale]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[benmccann]

There's some really good stuff here, but it's also just not possible to review given the size of the PR. Can you split it up into multiple smaller PRs? We've been trying to clear out the review queue, so I'm going to go ahead and close this given that I don't see any path forward on it in its current state. That doesn't mean that we don't appreciate this work, but it's just acknowledging the reality that no PR this big can be reviewed, which is why it's sat here for so long. If you'd like to discuss ways of splitting up the PR you can find us on Discord in the #contribute-to-svelte channel