Regression test for the zero-length-emit tile corruption (companion to #84) #85

Closed

StoneCypher wants to merge 1 commit from StoneCypher/view:test-tile-corruption-regression into main

StoneCypher commented

2026-07-06 02:46:34 +02:00

First-time contributor

Optional companion to #84 -- a regression test for the tile-tree corruption
that PR fixes. Kept as a separate PR so it can be adopted (or not)
independently of the fix.

Why it's a Node-side test, not a webtest

The corruption only reproduces under trusted input (a real drag).
Synthetic DOM events and programmatic dispatch({selection}) do not trigger
it, so an in-page webtest-* cannot catch it. This is a test-* (Node-side,
run by cm-runtests via Mocha) that drives a real pointer through Selenium's
Actions API. selenium-webdriver is already a dependency of
@marijn/testtool, and it needs a Chrome/chromedriver like the webtest suite.

What it checks

It builds a view with a whole-document mark (a lint diagnostic with no source
position) plus highlightActiveLine, performs a real drag-select + blank-line
click, and asserts the tile invariant:

state.doc.length - docView.tile.length

0 for a healthy tree; > 0 once a line's tile is dropped.

Status

Red against current main (the bug is still present) -- by design; this
is the reproduction.
Green with #84 applied.

So the intended order is: merge #84, then this. Merging this alone will (
correctly) fail CI until the fix lands.

Notes

The file carries // @ts-nocheck: Node builtins and selenium-webdriver
are untyped in this repo (no @types/node / @types/selenium-webdriver), so
the Node-side test is transpiled, not type-checked. It compiles cleanly
through cm-buildhelper as-is. Happy to type it properly instead if you'd
prefer adding those dev types.
The gesture is probabilistic per click (~50%); the test runs a few fresh
loads and stops at the first observed deficit, so it's fast when red and
runs the full budget (~30s) when green.

Optional companion to #84 -- a regression test for the tile-tree corruption that PR fixes. Kept as a separate PR so it can be adopted (or not) independently of the fix. ## Why it's a Node-side test, not a webtest The corruption only reproduces under **trusted** input (a real drag). Synthetic DOM events and programmatic `dispatch({selection})` do not trigger it, so an in-page `webtest-*` cannot catch it. This is a `test-*` (Node-side, run by `cm-runtests` via Mocha) that drives a real pointer through Selenium's Actions API. `selenium-webdriver` is already a dependency of `@marijn/testtool`, and it needs a Chrome/chromedriver like the webtest suite. ## What it checks It builds a view with a whole-document mark (a lint diagnostic with no source position) plus `highlightActiveLine`, performs a real drag-select + blank-line click, and asserts the tile invariant: ``` state.doc.length - docView.tile.length ``` 0 for a healthy tree; > 0 once a line's tile is dropped. ## Status - **Red against current `main`** (the bug is still present) -- by design; this is the reproduction. - **Green with #84 applied.** So the intended order is: merge #84, then this. Merging this alone will ( correctly) fail CI until the fix lands. ## Notes - The file carries `// @ts-nocheck`: Node builtins and `selenium-webdriver` are untyped in this repo (no `@types/node` / `@types/selenium-webdriver`), so the Node-side test is transpiled, not type-checked. It compiles cleanly through `cm-buildhelper` as-is. Happy to type it properly instead if you'd prefer adding those dev types. - The gesture is probabilistic per click (~50%); the test runs a few fresh loads and stops at the first observed deficit, so it's fast when red and runs the full budget (~30s) when green.

StoneCypher added 1 commit

2026-07-06 02:46:34 +02:00

Add a regression test for the empty-emit tile-tree corruption 3c39d83966

A Node-side test (run by cm-runtests via Mocha) that drives a real drag
through Selenium's Actions API against the built module. The corruption
only fires under trusted input, so it cannot be reproduced by an in-page
webtest (synthetic events and programmatic dispatch do not trigger it);
this drives real pointer input instead. selenium-webdriver is already a
dependency of @marijn/testtool.

It asserts the tile invariant state.doc.length - docView.tile.length:
0 for a healthy tree, > 0 once a line's tile is dropped. Currently fails
against main (which still has the bug) and passes with the openWidget fix
applied.

StoneCypher referenced this pull request

2026-07-06 04:05:25 +02:00

Fix tile-tree corruption from a zero-length emit under a document-wide mark #84