fix: don't transform main guard inside strings/comments in convert by vidigoat · Pull Request #10237 · marimo-team/marimo

vidigoat · 2026-07-19T07:29:32Z

This pull request was authored by a coding agent.

📝 Summary

marimo convert corrupts valid Python scripts when the text if __name__ == "__main__": appears inside a string literal or a comment.

_transform_main_blocks located the guard with a naive cell.code.split('if __name__ == "__main__":', 1), so any occurrence of that text — even inside a string or comment — was treated as a real guard. The cell was sliced mid-token and rewritten to def _main_(): …, producing syntactically invalid code that is then emitted as an app._unparsable_cell(...) blob.

This affects marimo convert script.py, marimo edit on a non-marimo .py file, and the server template API — both the plain-script and jupytext/pypercent paths.

🔍 Reproduction (before this PR)

from marimo._convert.converters import MarimoConvert

# valid Python — the guard text is just a string
src = 'GUARD = \'if __name__ == "__main__":\'\nprint(GUARD)\n'
print(MarimoConvert.from_non_marimo_python_script(src, aggressive=False).to_py())

Output on main slices the string literal and emits a mangled app._unparsable_cell(...). The same happens for a comment that mentions the guard.

🔧 Fix

Detect the guard via the AST: find a top-level ast.If whose test is __name__ == "__main__" (either operand order) with no else, and rewrite only that statement using line/column offsets. Cells that don't parse are left untouched (they're emitted as unparsable cells downstream anyway).

As a bonus this also fixes two cases that were silently not transformed before: the single-quoted form (if __name__ == '__main__':) and the reversed comparison (if "__main__" == __name__:).

✅ Tests

Added TestTransformMainBlocks (6 cases: guard-in-string, guard-in-comment, single quotes, reversed comparison, and that a real guard still transforms).
uv run --python 3.13 --group test pytest tests/_convert tests/_cli/test_cli_convert.py → 363 passed, 2 skipped (env), 1 xfail (pre-existing, unrelated).
ruff check, ruff format --check, and mypy all clean on the changed files.

📋 Checklist

I have read the CLA
For new features, I have added tests
I have run the tests and they pass

marimo convert located the `if __name__ == "__main__":` block with a naive string search, so the text appearing inside a string literal or comment was treated as a real guard. The cell was then sliced mid-token and rewritten, producing invalid code that was emitted as an unparsable cell. Detect the guard via the AST instead: find a top-level `if` whose test is `__name__ == "__main__"` (in either order) with no `else`, and rewrite only that. Cells that don't parse are left untouched. This also fixes the single-quoted and reversed-comparison forms, which were silently not transformed before.

vercel · 2026-07-19T07:29:36Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jul 19, 2026 7:30am

github-actions · 2026-07-19T07:29:41Z

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

vidigoat · 2026-07-19T07:41:41Z

I have read the CLA Document and I hereby sign the CLA

dmadisetti · 2026-07-20T21:59:55Z

+ # Absolute offset of each line start (ast linenos are 1-based and
+ # count physical "\n"-separated lines).
+ line_starts = [0]
+ for line in cell.code.split("\n"):
+ line_starts.append(line_starts[-1] + len(line) + 1)
+
+ start = line_starts[guard.lineno - 1] + guard.col_offset
+ test_end_lineno = guard.test.end_lineno or guard.lineno
+ test_end_col = guard.test.end_col_offset or 0
+ test_end = line_starts[test_end_lineno - 1] + test_end_col
+ # The colon that closes the `if` header is the first ":" after the
+ # end of the test expression (comments cannot appear before it).
+ colon = cell.code.index(":", test_end)
+
+ before_main = cell.code[:start].strip()
+
+ # replace the if __name__ == "__main__": with def _main_():
+ main_block = "def _main_():" + cell.code[colon + 1 :] + "\n\n_main_()"
+
+ if before_main:
+ cell.code = before_main + "\n\n" + main_block
+ else:
+ cell.code = main_block


I think an AST transform may be better than this. I'm wondering if replacing with def _main_ makes sense either.

I reckon if mo.app_meta().mode == "script" almost makes more sense?

cc @manzt who last touched this

vidigoat · 2026-07-21T10:02:52Z

Thanks @dmadisetti! Two things:

On the AST point — the PR does use AST to find the guard (that's the actual fix here; the old code string-matched if __name__ == "__main__": and mangled it whenever that text showed up inside a string literal or comment). I deliberately kept the replacement as a string splice rather than ast.unparse, because unparsing the whole cell would drop comments and reformat the user's code — which seems bad for a convert tool that should round-trip their source as faithfully as possible. Happy to go fully-AST if you'd prefer, though.

On def _main_() vs mo.app_meta().mode == "script" — I think you're right that the latter is a better fit: it preserves the original "only run when executed directly" semantic instead of wrapping the body in a function and calling it. I kept def _main_() for now, but I'm glad to rework toward if mo.app_meta().mode == "script": if that's the direction you and @manzt want.

It's still a draft, so happy to settle the approach first before polishing.

dmadisetti · 2026-07-21T22:01:09Z

unparsing the whole cell would drop comments and reformat the user's code

Agreed here. Cool I think this makes sense

vidigoat · 2026-07-22T09:45:15Z

Thanks @dmadisetti! Glad the string-splice reasoning makes sense. The one open question is the target: keep def _main_(): ... _main_(), or switch the guard body to run under if mo.app_meta().mode == "script": as you suggested (which arguably preserves the original "only when run as a script" semantic more faithfully). I'm happy to go either way — just let me know which you and @manzt prefer and I'll finalize and take it out of draft.

@vidigoat

Avoids mangling cells when the guard text appears inside a string or comment. Also handles single-quoted and reversed comparisons. Based on work by @vidigoat in marimo-team#10237. Co-authored-by: Vidit Patankar <vidit.patankar16@gmail.com>

vercel Bot deployed to Preview July 19, 2026 07:30 View deployment

Light2Dark requested a review from dmadisetti July 20, 2026 04:35

dmadisetti reviewed Jul 20, 2026

View reviewed changes

Bartok9 mentioned this pull request Jul 22, 2026

fix(convert): AST-based __main__ guard (salvage of #10237, credit @vidigoat) #10284

Closed

3 tasks

dmadisetti mentioned this pull request Jul 22, 2026

docs(convert): note how __main__ guards are rewritten #10286

Closed

2 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: don't transform main guard inside strings/comments in convert#10237

fix: don't transform main guard inside strings/comments in convert#10237
vidigoat wants to merge 1 commit into
marimo-team:mainfrom
vidigoat:fix/convert-main-guard-in-string

vidigoat commented Jul 19, 2026

Uh oh!

vercel Bot commented Jul 19, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jul 19, 2026 •

edited

Loading

Uh oh!

vidigoat commented Jul 19, 2026

Uh oh!

dmadisetti Jul 20, 2026

Uh oh!

dmadisetti Jul 20, 2026

Uh oh!

vidigoat commented Jul 21, 2026

Uh oh!

dmadisetti commented Jul 21, 2026

Uh oh!

vidigoat commented Jul 22, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

vidigoat commented Jul 19, 2026

📝 Summary

🔍 Reproduction (before this PR)

🔧 Fix

✅ Tests

📋 Checklist

Uh oh!

vercel Bot commented Jul 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Jul 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vidigoat commented Jul 19, 2026

Uh oh!

dmadisetti Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

dmadisetti Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

vidigoat commented Jul 21, 2026

Uh oh!

dmadisetti commented Jul 21, 2026

Uh oh!

vidigoat commented Jul 22, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

vercel Bot commented Jul 19, 2026 •

edited

Loading

github-actions Bot commented Jul 19, 2026 •

edited

Loading