Armando Pellegrini 59e16e0187 Fix State.dispose() not invoking transformer dispose for live cells (#1826) ...

`Transformer.Definition.dispose` is documented as "automatically called
on deleting an object," but `State.dispose()` only disposed its own event
subjects and action manager — it never iterated still-live cells to call
their per-transformer dispose. Cells holding GL buffers, mesh data, etc.
only had their dispose fired on explicit deletion (e.g. `clear()`), so
any consumer that called `plugin.dispose()` without first awaiting
`plugin.clear()` retained the callback chain, the GL buffers it points
at, and any closures captured by it.

In a long-running single-page app where the user navigates between
routes that mount/unmount a Mol* viewer, this leaked roughly 25–50 MB
of process RSS per cycle even with `plugin.dispose()` correctly called.
A 20-cycle E2E mount/unmount harness on a 1AKE structure measured a
+541 MB RSS / +266 MB JS-heap delta in the unconditional-`dispose()`
case; calling `await plugin.clear()` before `plugin.dispose()` halved
the residual leak, confirming the per-cell dispose path was missing on
the unconditional `dispose()` route.

This change walks the cell tree once (post-order via the existing
`StateTree.doPostOrder` helper) and invokes the per-transformer dispose
for every still-live cell, swallowing+warning on errors so a single
faulty transformer can't prevent siblings from cleaning up. The
existing per-cell `dispose` helper is reused for consistency with
`updateNode`/`findDeletes` semantics.

Tests cover: chained transformers, sibling subtrees, throwing-dispose
isolation, and transformers without a dispose definition.

Also adds `useDefineForClassFields: false` to the jest esbuild
transform so tests can construct `State` (the `TransientTree` parameter
property + class field pattern relies on legacy class-field semantics,
which `tsc` honors via `target: es2018` but esbuild's default `esnext`
target does not).

Fixes #1825

Co-authored-by: Armando Pellegrini <tech.tools@boltz.bio>

2026-05-09 22:17:38 +02:00

7.1 KiB

Raw Permalink Blame History

View Raw