ISSUE DESCRIPTION marimo edit --watch messes with merge conflict markers Describe the…
- User
ISSUE DESCRIPTION
marimo edit --watch messes with merge conflict markers
Describe the bug
When cherry-picking commits, or
git stash poping, sometimes merge conflicts occur.If a notebook is open using
marimo edit --watchduring that time, it parses the git conflict markers asapp._unparsable_cell. This gets in the way of resolving the conflicts using conflict resolution tools likegit-mediateSuggestion: Avoid updating the notebook if the file contains conflict markers, recognizable by
<<<<<<<at the beginning of a line.Will you submit a PR?
- Yes
Environment
{ "marimo": "0.23.6", "editable": false, "location": "/Users/yairchu/dev/soundradix/Drum3D/.venv/lib/python3.13/site-packages/marimo", "OS": "Darwin", "OS Version": "25.2.0", "Processor": "arm", "Python Version": "3.13.1", "Locale": "C/en_US", "Binaries": { "Browser": "140.0.7339.134", "Node": "v23.11.0", "uv": "0.7.12 (dc3fd4647 2025-06-06)" }, "Dependencies": { "click": "8.4.0", "docutils": "0.22.4", "itsdangerous": "2.2.0", "jedi": "0.19.2", "markdown": "3.10.2", "narwhals": "2.21.2", "packaging": "26.2", "psutil": "7.2.2", "pygments": "2.20.0", "pymdown-extensions": "10.21.3", "pyyaml": "6.0.3", "starlette": "1.0.0", "tomlkit": "0.15.0", "typing-extensions": "4.15.0", "uvicorn": "0.47.0", "websockets": "16.0" }, "Optional Dependencies": { "altair": "6.1.0", "anywidget": "0.11.0", "loro": "1.10.3", "pandas": "3.0.3" }, "Experimental Flags": {} }Code to reproduce
No response
ADDITIONAL ADVICE
Since you are given a git repository, you can use git commands to simplify your work. For example, if you made a mistake and want to revert your changes, you can use
git checkout <file>to restore the file to its original state. You can also reset all changes in the repository usinggit reset --hardcommand. Additionally, you can usegit stashandgit stash popto temporarily save your changes and restore them later. Finally, you can see the changes that you've made (which will be submitted when you callsubmit) usinggit statusorgit diffcommands. However you don't need to usegit addorgit commitbefore submitting your changes. In fact,submitonly submits currently unstaged changes because it usesgit diffto get the changes that need to be submitted.CHECKLIST
Before submitting your solution, please go over the following checklist and make sure you've done everything:
- If an error was reported in the issue description, I have successfully reproduced it.
- If an error was reported in the issue description, I have confirmed that my fix resolves the error.
- I have thought about edge cases and made sure my fix handles them as well.
- I have run existing tests in the repository that might have been affected by the change I made and confirmed that they pass. I want you to list every bullet from this checklist and write your reasoning for why you think you did it or didn't need to.
Repository has been uploaded and your shell is currently at the repository root. Time to solve the issue!
(Current directory: /marimo, current file: none) bash-$
- Assistant
I’ll trace the
--watchfile-change path, identify where notebook source is read and reparsed, then add a guard at the narrowest appropriate layer so conflict-bearing files are ignored without affecting normal parsing. I’ll also locate focused tests and reproduce the behavior before and after the change. - Assistant
ripgrepis unavailable, so I’ll use recursivegrepand targeted directory inspection to find the watcher implementation and related tests. - Assistant
The relevant path appears to be
SessionFileWatcherExtension, which is specific to live session reloads (unlike export watchers or the generic filesystem observer). I’ll inspect that extension, its callback in session management, and its focused test suite to decide whether the conflict check belongs before callback dispatch or during reload. - Assistant
The extension currently forwards every change unconditionally. Before editing, I need to inspect the actual reload callback and notebook file manager APIs, because reading through an existing abstraction may handle encoding/errors consistently and may be the better place for the guard.
- Assistant
I found a dedicated
FileChangeCoordinatorbehind the session manager callback. I’ll inspect it in full, since it likely owns reload validation, locking, and error handling; placing the marker detection there could cover all live reload entry points (including test-triggered changes) while preserving generic watcher behavior. - Assistant
FileChangeCoordinatoris the right semantic boundary: it already suppresses self-writes and catches invalid reloads, and both watcher-driven and explicit file-change handling pass through it. The guard must happen beforereload(), otherwise the in-memory notebook is already mutated into unparsable cells. I’ll inspectAppFileManager.read_file()and coordinator tests/mocks to implement the check without introducing redundant or inconsistent file I/O. - Assistant
The tests confirm current syntax-error fallback intentionally reloads malformed cells, so the fix should be narrowly limited to lines beginning exactly with
<<<<<<<, as requested, rather than suppressing all parse errors. I’ll inspect the file manager’s read semantics and the rest of coordinator error tests, then reproduce the conflict mutation with a small temporary script. - Assistant
read_file()uses the configured storage backend, so it is preferable to directPath.read_text()and works for non-default storage. I’ll first run a temporary reproduction that records the cell manager before and after a coordinator reload of a file containing a standard conflict block; this should demonstrate the reported mutation and establish the expected post-fix invariant. - Assistant
The initial reproduction failed only because Python sets
sys.pathfrom the temporary script directory. I’ll rerun it with the repository root onPYTHONPATH; no code changes have been made yet. - Assistant
The base interpreter lacks repository dependencies (
msgspec). The project has a lockeduvenvironment, so I’ll useuv runfor the reproduction and subsequent focused tests rather than altering the environment manually. - Assistant
The current branch already catches this particular boundary-marker case as a syntax error, but that is incidental parser behavior and does not guarantee conflict blocks embedded where the best-effort scanner can recover. It also logs the event as an error rather than intentionally recognizing a merge conflict. I’ll add an explicit pre-reload check in the coordinator using
read_file()andsplitlines(), matching only<<<<<<<at column zero. The intentional skip will returnhandled=Falsewith no error and leave both the document and reload strategy untouched. Read failures will remain handled by the existing reload exception path. - Assistant
The guard is in place and the diff is clean. I’ll now rerun the reproduction to verify it is an intentional no-op, then run the complete focused coordinator test module. I’ll also run static formatting/lint checks available through the project tooling afterward.
- Assistant
The reproduction now confirms the conflict-bearing file is ignored and in-memory cells stay unchanged; all 23 coordinator tests pass, including normal external edits and syntax-error fallback. I’ll review edge behavior with custom storage/mocks and project style conventions. In particular, the check intentionally ignores only column-zero markers, permits ordinary
<<<<<<<substrings and indented lines, and uses universal line splitting (including CRLF). - Assistant
The configured line length is 79, and the
any(...)line exceeds it. I’ll run Ruff’s formatter on the changed source and then Ruff check on that file, followed by the watcher integration tests because they exercise the upstream callback path. - Assistant
The implementation is formatted, lint-clean, and both the coordinator (23 tests) and watcher integration (9 tests) suites pass. To validate the precise historical failure mode rather than only the boundary-marker variant, I’ll probe several standard conflict placements by calling
AppFileManager.reload()directly (bypassing the new guard). This is temporary diagnostic code and will not modify repository tests. - Assistant
The direct probe reproduces the reported bug exactly: when markers occur inside a cell body,
reload()succeeds and replaces the cell code with conflict text (the best-effort unparsable representation path). I’ll update the temporary reproduction to use that placement and verify two key lifecycle properties through the coordinator: the conflicted intermediate state is skipped, and a later resolved file change reloads normally. - Assistant
The focused suites still pass (32 total). The temporary lifecycle script reached and passed all conflict-skip assertions, then failed only because my expected normalized cell string omitted indentation; this is an overly exact diagnostic assertion, not a product failure. I’ll correct it to assert semantic content (
x = 2, no conflict marker), rerun, and then perform final status/diff checks.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- gpt-5.6-solCompare side by side →
- gpt-5.6-solCompare side by side →
- gpt-5.6-solCompare side by side →