6 Commits

Author SHA1 Message Date
Andrey Antukh
69c33c4877
♻️ Wait on signals instead of event-loop turns on frontend tests (#12986)
* ♻️ Wait on signals instead of event-loop turns

`wait-for` counted its budget in event-loop turns (50 x `setTimeout 0`,
about 54 ms idle), so a delivery that travelled through the RxJS async
scheduler could arrive after the wait gave up. It now accepts a
`:signal`: the predicate is re-checked on every emission and the wait
resolves on the event that changes the state. The ceiling (5000 ms)
only bounds failure when a signal is given, and stays a poll budget for
the waits that have nothing to listen to.

`wait-for-store` wraps the store's input stream, which notifies its
subscribers after applying the update. The five explicit `:max-ticks`
sites move to `:timeout-ms` in the same change, so the old option never
lives in a state where destructuring silently ignores it.

New `frontend-tests.helpers.async-test` pins the contract: a late
delivery resolves, a store signal resolves on the event without a poll,
an already-true predicate does not yield, a never-true predicate records
one failure within the ceiling, and a resolved wait stops listening.

AI-assisted-by: deepseek-v4.1-flash

* ♻️ Wait for the persistence requests on their own stream

The persistence tests waited for `(= N (count @requests))` by polling the
atom at every event-loop turn. The mocked transport now pushes each
request into a `requests$` subject, and those 59 waits pass it as
`:signal`, so they resolve on the request that makes the count true. The
two predicates that mix the request count with the store state merge
both signals.

The wait that flaked (`the other file's edit is sent next`, in the
rollback test) is one of these: it no longer depends on a turn budget,
and it cannot lose the delivery.

AI-assisted-by: deepseek-v4.1-flash

* ♻️ Wait for the store event in the store backed tests

153 waits whose predicate reads the store now go through
`wait-for-store`, so they resolve on the store event that changes the
state instead of polling at every event-loop turn. The count is 153, not
the 162 the plan estimated: nine of them read the store through a
helper (`fill-color`, `shape-name`, `copy-children`) rather than
`@store`. The five sites that carried an explicit ceiling drop it, since
with an event the ceiling only bounds failure.

Resolving on the event exposes the places where the poll was really a
pause for a side effect to advance:

- `tokens-source-test` emitted `unlink-file-from-library`, a watch that
  chains a second `rp/cmd!` after an async hop. The store signal fires
  before the chain runs (the event itself dissocs the library), so the
  mock scope closed in flight and the second call reached the real HTTP
  layer. The test now waits for that second call, through the mock.
- `persistence-rollback-test` drives the persistence pipeline step by
  step; two of its waits are the pause before the next emit, not an
  effect, so they keep the poll.

AI-assisted-by: deepseek-v4.1-flash

* 📚 Document the wait contract and what still polls

`mem:frontend/testing` described `wait-for` as an immediate check plus a
bounded poll. It now waits on a signal, and the ceiling only bounds
failure; `wait-for-store` and the persistence `requests$` subject are
how a wait resolves on the event that changes the state. The waits with
nothing to listen to keep the poll, and the memory names them.

The two traps that cost a red suite are recorded too: a wait that means
"the side effect advanced" has no event and keeps the poll, and a
store-state wait can close the mock scope before the flow it observes
ends, so the chain's last RPC call has to be awaited through the mock.

AI-assisted-by: deepseek-v4.1-flash

* ♻️ Give the wait wrapper a single signal

`wait-for-store` built its options with `assoc`, so it replaced the
`:signal` a caller passed. The two waits whose predicate reads both the
store and the request counter relied on a merged signal that was being
discarded; they now pass it to `wait-for` themselves, and the wrapper
takes no options. That merge was the safety net for a request counted
after the store event, and it was inert.

The helper also reports why a wait failed (ceiling, ended signal, or an
error with its cause) instead of calling all three a timeout, counts
the predicate evaluations in the late-delivery case to prove the signal
path does not poll, reads the clock through `app.common.time` in the
helper and the test, and the test's turn probe cancels its timer. The
copyright header of `helpers/async.cljs` stays as the previous commit
left it, matching the majority spelling.

AI-assisted-by: deepseek-v4.1-flash

* ♻️ Give the cycle counter a signal too

Two waits in the retry tests read the store and the `cycles` atom, which
the `slow-cycle-timer` double fills inside the persistence watch. They
passed only the store signal, so a counter change could not wake them:
the same latent race the previous commit removed from the request count.
The double now also emits each new cycle into a `cycles$` subject, and
both waits pass the merged signal to `wait-for`.

The public `wait-for` docstring names the three failure reasons, and the
message for a failed signal bounds the cause text: the exception message
when there is one, otherwise the printed value inside a try, cut at 500
characters.

AI-assisted-by: deepseek-v4.1-flash

* ♻️ Stop the error handlers flooding the test log

Four tests in `main-errors-test` run a real error handler end to end.
Those handlers print the cause with `ex/print-throwable` before
notifying, which is what a developer wants in the app; in a test it only
writes console noise. One of them replays 10,000 offline errors, so a run
carried 10,006 dumps: 190,010 of its 192,682 lines, 13.8 MB, well past
the ~4 MB where GitHub truncates a step log and takes the summary and any
failing assertion with it.

The tests stub `ex/print-throwable` with a no-op for the duration of
their mock scope. The handlers still run, the assertions do not change,
and the 10,000-error loop still proves the governor emits once. A full
run now logs 2,583 lines and 384 KB. The single block that remains comes
from the re-entrancy guard, which prints its own notice on purpose.

AI-assisted-by: deepseek-v4.1-flash

* ♻️ Silence the deliberate WASM panic in its test

`get-svg-markup` logs a WASM panic with `console.error` before returning
empty, and `get-svg-markup-swallows-wasm-errors` sends two panics through
it on purpose. That was two 35-line dumps in which the exception printed
its own `data`, `arr` and protocol masks, and it reads like a failure that
the suite is hiding.

The test replaces `(.-error js/console)` for its body and restores it in a
`finally`. The assertions and the helper stay as they are: the same
panic answered with a different stack would still show up, because only
this test silences the console.

AI-assisted-by: deepseek-v4.1-flash

* 🐛 Use real timer for wait-poll ceiling

wait-poll read the app clock (ct/now), so tests freezing that clock
never reached the deadline and hung instead of failing. Track expiry
with a real setTimeout flag, pass timeout-ms through wait-signal and
wait-for, and cover the frozen-clock case with a regression test.

AI-assisted-by: muse-spark-1.3-contributor
2026-11-13 09:18:43 +01:00
alonso.torres
7eacbec245 🐛 Keep comments, focus, thumbnails and text in step with a rollback
A rollback reverts the local data with a rebase, which is not a commit,
so the parts of the workspace that follow commits missed it:

- Comment pins moved in the reverted action stayed where they were, on
  screen and on the server, and pins on a board the rollback removed
  kept pointing to it. They go back to the reverted position, and pins
  on a removed board move to the page root.
- Focus mode stayed on with no shape left when the rollback removed the
  focused ones, and the next shape drawn turned it on again. It ends
  when its last shape goes, and the selection drops removed shapes.
- Component thumbnails kept the rejected look. The thumbnails of the
  reverted mains are rendered and uploaded again, and an upload whose
  shapes changed during the render is skipped.
- Under WASM, an open text editor reset to the reverted content but the
  canvas kept the old layout. It is laid out again.
- A library sync whose save failed showed no update notice again once
  the release undid it. The notice is checked again after that
  rollback.
2026-11-05 23:59:32 +01:00
alonso.torres
61bf66342b ✨ Tell repairs, damaged data and other files apart in save notices
The notice of a rejected save always said the same thing. It did not
name the file when the user had moved to another one, gave no hint when
the page held data that was already invalid, and replaced the notice of
a backend repair still on screen, since the store holds one toast.

The backend flags a rejection whose first commit touches pages or
components that failed validation before it (:preexisting-damage). The
rejection notice names a file other than the open one, says when the
data was damaged before, and folds into a repair notice still on
screen; a later repair notice folds a rejection notice in the same way.
The notice says that the latest changes were undone, no longer that the
rest of the work is saved.
2026-11-05 23:59:32 +01:00
alonso.torres
302f83f072 🐛 Drop the unsaved edits made on a version a peer restored
When a peer restores a version, the file reloads to it. Edits made on
the replaced version stayed queued: their save failed for good and
locked the workspace, and a rollback or a rebase replayed them on the
restored data.

Loading another version forgets the queued, buffered and saved commits
made on the old one. A save of such a commit, in flight or answered, is
dropped instead of failing or being rolled back, and one save carries
the edits of one version only.
2026-11-05 23:59:32 +01:00
alonso.torres
7538d2c187 ✨ Tell plugins and the MCP agent when changes are rolled back
The backend rejects a save after the call that made the changes has
returned, and the workspace then undoes them. A plugin, or the MCP
agent, went on working with shapes that were no longer there and never
learned why.

The rollback event now names the files that lost commits and the pages
and page shapes their changes touched, and plugins get it as the new
`contentrollback` event (`ContentRollback` type) through a subscription
to the store stream, so no rollback is dropped by debouncing.

The MCP plugin keeps the rollbacks it is told about and sends them with
its next task response, failed or not. The server collects them per tool
call and puts a warning first in that tool's response, naming what was
affected; the instructions tell the agent to read the pages again and
redo what is missing.
2026-11-05 23:59:32 +01:00
alonso.torres
94c05589cf 🐛 Keep the UI in step with a rollback
A rollback can remove or change shapes the UI still refers to.

- The selection drops shapes that are gone and the color picker that
  edited them closes.
- The undo history is cut at the reverted action; it stays when no
  entry holds a reverted change. Undo and redo commits carry real undo
  changes and the history index they moved from, so a reverted Ctrl+Z
  puts the change and the index back.
- An open path edition whose shape the rollback changed closes without
  saving the editor's stale copy. The v2 text editor loads the reverted
  text again; the WASM editor already showed it.
- The Alt measurements look the hovered shape up in the current data.
- Plugin setters, resize() and rotate() on a removed shape throw
  "Shape not found", whatever the plugin's throwValidationErrors flag.
2026-11-05 23:59:32 +01:00