… handling (#9610)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

…adManager (#9590)

…ils (#9596)

show cell index in dependency minimap

<img width="651" height="350" alt="Screenshot 2026-05-20 at 5 06 55 PM"
src="https://github.com/user-attachments/assets/d15b0545-f9fb-4b23-a7e6-5a4bb8d35fa7"
/>
<img width="655" height="360" alt="Screenshot 2026-05-20 at 5 06 49 PM"
src="https://github.com/user-attachments/assets/3e86d2b0-b755-43d2-9a3a-8cc343973394"
/>

Adds behavior tests for four untested surfaces of the executor refactor,
and includes the one production-fix uncovered while planning.

Production fix (``marimo/_runtime/executor/executor.py``): let
``asyncio.CancelledError`` propagate unwrapped from
``DefaultExecutor.execute_cell{,_async}``. Wrapping it in
``MarimoRuntimeException`` defeated the bare-``CancelledError`` check
in ``Runner._finalize_run_result`` and prevented
``runner.interrupted`` from flipping on async SIGINT.

Tests:
- ``StrictLifecycle.setup`` ``Skip(result=...)`` path: 5 cases
  covering undefined ref, ref-before-def, graph-resolved blamed cell,
  private-var ``unmangle_local`` fallback, and the state invariant
  that the Skip early-return must not mutate globals or stash a
  backup (teardown then no-ops).
- ``unwrap_user_exception``: 6 cases, including the
  ``NameError.name is None`` guard. Construction note: bare
  ``NameError("name 'x' is not defined")`` does NOT set ``.name == 'x'``
  — only interpreter-raised NameErrors do. Tests set ``.name`` explicitly.
- Plugin executor dispatch via ``Runner``: registers a sentinel
  factory against ``_EXECUTOR_REGISTRY`` (fully isolated — both
  ``_plugins`` and ``names`` monkeypatched so installed third-party
  entry points can't shadow), asserts the runner runs the sentinel.
- SIGINT and ``runner.interrupted``: deterministic ``_cancel_on_sigint``
  tests via ``signal.signal``/``signal.getsignal`` monkeypatching
  (real signals are unsafe in pytest — the wrapper chains to the
  ``previously installed`` handler, which in tests is pytest's, not
  marimo's), plus the executor-level CancelledError-propagation test
  and runner-level tests for both sync ``MarimoInterrupt`` and async
  cancellation flipping the flag.

Resolves cubic's "this should be tested" comment on
``evaluator.py:97``.

- Inline ``Evaluator(...)`` at the two remaining ``build_evaluator``
  call sites (``by_kwargs._new_evaluator`` and
  ``AppScriptRunner.__init__``) so the helper removal on
  ``dm/executor-impl`` doesn't leave dangling imports here.
- Extract ``_strip_frame(e, count=1)`` util in ``executor.py`` for
  the executor-frame elision logic. Iterates ``count`` times and
  stops if the traceback runs out, never stripping the only frame
  we have — fixes the subtle issue where ``tb_next`` being ``None``
  would set ``__traceback__`` to ``None`` outright.
- ``_classify`` in ``by_kwargs`` now short-circuits on a bare
  ``MarimoStopError`` before the ``MarimoRuntimeException`` unwrap
  branch — any caller bypassing the default wrap (e.g. a custom
  ``Executor`` that raises ``MarimoStopError`` directly) still gets
  stop-control-flow handling.

…#9630)

## 📝 Summary

Replaces `dict[str, Any]` in the executor pipeline with `MutableGlobals`
(alias), and `Globals` (`Mapping[str, Any]`).

Layering on top to prevent rebase churn

**This pull request was authored by a coding agent.**

Fixes #9628.

With `auto_reload` set to `lazy` or `autorun`, every cell run was
calling `ModuleReloader.check(sys.modules, reload=True)`, which iterates
*all* of `sys.modules` and does `os.stat` on each entry. With ~1000
modules in scope (typical), that adds 16–80ms per cell — compounded
across the dozen cells re-running on a UI interaction it becomes a >1s
lag.

This change adds an opt-in `skip_non_user_modules=True` flag on
`ModuleReloader.check`. When set, stdlib and site-packages module names
are recorded in a persistent skip set (classified by `sysconfig`
prefixes) and short-circuited on subsequent calls.

`AutoreloadManager.cell_scope` (the hot per-cell path) opts in. The
background `ModuleWatcher` keeps the default behavior and continues to
scan every module on its 1s loop, so edits inside an installed package
are still detected — just at watcher latency rather than cell-entry
latency. Editable installs (`pip install -e .`, `uv add --editable`)
have `__file__` outside site-packages, so they are correctly classified
as user code and reload with no latency change.

### Benchmark

Driving `ModuleReloader.check()` directly, 200 iterations post-warmup.
Issue-shaped workload: ~2.5k modules (heavy stdlib + numpy/pandas/etc.)
+ 5 user files in a tmp dir.

| path | median | p95 |
|------|-------:|----:|
| before | 4.88 ms | 6.15 ms |
| after  | 0.91 ms | 1.01 ms |

**~4 ms saved per cell run, 5.4× median speedup.**

Scale curve (median µs, varying user-module count):

| user mods | sys.modules | before | after | speedup |
|----------:|------------:|-------:|------:|--------:|
| 0   | 2514 |  5037 |  873 | 5.8× |
| 5   | 2519 |  5245 |  802 | 6.5× |
| 25  | 2539 |  6082 | 1693 | 3.6× |
| 100 | 2614 |  8342 | 4421 | 1.9× |
| 500 | 3014 | 12489 | 8398 | 1.5× |

The win narrows as user-code grows, by design: the optimization only
filters out non-user-code.

## 📝 Summary

<!--
If this PR closes any issues, list them here by number (e.g., Closes
#123).

Detail the specific changes made in this pull request. Explain the
problem addressed and how it was resolved. If applicable, provide before
and after comparisons, screenshots, or any relevant details to help
reviewers understand the changes easily.
-->
1. Fix playwright test
2. For a notebook with very few cells, we didn't stretch the slide area
fully. This fixes that. See the snapshot picture for reference.

## 📋 Pre-Review Checklist
<!-- These checks need to be completed before a PR is reviewed -->

- [ ] For large changes, or changes that affect the public API: this
change was discussed or approved through an issue, on
[Discord](https://marimo.io/discord?ref=pr), or the community
[discussions](https://github.com/marimo-team/marimo/discussions) (Please
provide a link if applicable).
- [x] Any AI generated code has been reviewed line-by-line by the human
PR author, who stands by it.
- [ ] Video or media evidence is provided for any visual changes
(optional). <!-- PR is more likely to be merged if evidence is provided
for changes made -->

## ✅ Merge Checklist

- [x] I have read the [contributor
guidelines](https://github.com/marimo-team/marimo/blob/main/CONTRIBUTING.md).
- [ ] Documentation has been updated where applicable, including
docstrings for API changes.
- [x] Tests have been added for the changes made.

…ST) (#9645)

Our docstrings are in Markdown, not RST.

This PR adds a precommit that enforces markdown-style inline code
(single backticks), and a note in AGENTS.md telling agents about this
convention. The PR also fixes all offending instances of double
backticks.

This convention matters because it affects how docstrings are
parsed/rendered in our doc. It also matters for just internal style
consistency.

---------

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

## 📝 Summary

<!--
If this PR closes any issues, list them here by number (e.g., Closes
#123).

Detail the specific changes made in this pull request. Explain the
problem addressed and how it was resolved. If applicable, provide before
and after comparisons, screenshots, or any relevant details to help
reviewers understand the changes easily.
-->
- Pressing Enter with no input in marimo's stdin box was silently
swallowed by the frontend, so any CLI calling `input("Press Enter to
continue:")` (e.g. `logfire`, `click.prompt`, `rich.prompt`, `pdb`)
would hang forever.
- Once unblocked, a second bug surfaced: `builtins.input()` raised
`EOFError` on blank submissions because `ThreadSafeStdin.readline()`
returned `""` — which CPython's `input()` interprets as EOF rather than
a blank line.

### Fix

**Frontend** (`ConsoleOutput.tsx`)
- `StdInput` now submits on Enter even when empty (history skip
preserved).
- `StdInputWithResponse` renders past responses with a `>` chevron,
sky-11 color, and an `(empty)` placeholder so user-typed responses are
visually distinct from stdout — even when the prompt half is empty
(which happens when CLIs `print()` the prompt themselves before calling
`input()`).

<img width="501" height="191" alt="image"
src="https://github.com/user-attachments/assets/59999aad-09fe-4174-9408-b009981130be"
/>

**Backend** (`marimo/_messaging/types.py`, `streams.py`,
`_pyodide/streams.py`)
- `Stdin.readline()` now appends `"\n"` so blank submissions surface as
`"\n"` rather than `""`. `readlines()` simplified to `[readline()]`.
- Lifted `readline`/`readlines` to the `Stdin` base class — both
subclasses had identical implementations.
- `_readline_with_prompt` unchanged: still returns the bare response, so
`input_override` continues to satisfy Python's `input()` contract.

## 📋 Pre-Review Checklist
<!-- These checks need to be completed before a PR is reviewed -->

- [x] For large changes, or changes that affect the public API: this
change was discussed or approved through an issue, on
[Discord](https://marimo.io/discord?ref=pr), or the community
[discussions](https://github.com/marimo-team/marimo/discussions) (Please
provide a link if applicable).
- [x] Any AI generated code has been reviewed line-by-line by the human
PR author, who stands by it.
- [x] Video or media evidence is provided for any visual changes
(optional). <!-- PR is more likely to be merged if evidence is provided
for changes made -->

## ✅ Merge Checklist

- [x] I have read the [contributor
guidelines](https://github.com/marimo-team/marimo/blob/main/CONTRIBUTING.md).
- [ ] Documentation has been updated where applicable, including
docstrings for API changes.
- [x] Tests have been added for the changes made.

…s-origin iframes (#9649)

#9636)

Removed the section on adding a formatter to the marimo repo. It's
better if folks make their own viewers for their own objects.