DIAGNOSTIC (do not merge): which code cells does thebe-lite actually execute? - #65
Draft
mmcky wants to merge 4 commits into
Draft
DIAGNOSTIC (do not merge): which code cells does thebe-lite actually execute?#65mmcky wants to merge 4 commits into
mmcky wants to merge 4 commits into
Conversation
… executes Temporary page to settle #64 by measurement rather than inference. It exists to make a silent failure visible. The problem it solves: when a cell does not execute under thebe-lite there is no error and no output — and for an assignment cell, "no output" is also the CORRECT result. Those two states are indistinguishable by eye, which is what made this bug take several wrong turns to pin down. So every cell here appends a marker to a list, and a final top-level REPORT cell prints which markers are present and which are MISSING. One click on that cell names every cell that silently did nothing. Positions probed, all in the same document: A top level - control, must work B top level, `%%file` magic - tests the MAGIC independently of nesting C top level, reads B's file back - did %%file actually write? D inside a gated {exercise-start} - the position that fails in short_path E inside a gated {solution-start} - the position that WORKS in short_path, but only because a malformed directive spills it to root level F inside a plain {note} admonition - is it exercises, or nesting generally? B and C are the ones that matter for the fix: `%%file` under thebe-lite has never actually been tested. It was assumed broken, and that assumption was wrong — the cell it was tried in was nested and never ran at all. If B/C pass at top level, `%%file` is viable and lecture-wasm can converge with the five sibling repos instead of carrying a divergent string-based helper forever. Not for merge. Delete the branch once #64 has its answer.
❌ Deploy Preview for beautiful-dodol-cb9543 failed.
|
mmcky
added a commit
that referenced
this pull request
Aug 12, 2026
BLOCKED on #64 — do not merge yet. See that issue and the diagnostic in #65. This PR is the intended END STATE; what is unproven is whether this runtime can run it. This lecture executes in the reader's browser (`lite: true`) and was the only repo in the family fetching this graph across repos on every page view: import requests file_url = "https://raw.githubusercontent.com/QuantEcon/lecture-python-intro/main/lectures/graph.txt" graph_file_response = requests.get(file_url) Every sibling — lecture-python-intro, lecture-dp, lecture-jax, lecture-intro.zh-cn and test-actions-lecture-intro — writes the data with a `%%file graph.txt` cell and reads it back with `open(in_file)`. This repo now does the same, so all six are identical on this exercise, and the divergent string-splitting helper `map_graph_to_distance_matrix(in_file_data)` returns to the canonical filename form. It also retires a cross-repo dependency: that `requests.get` was the only thing reading lecture-python-intro's committed `lectures/graph.txt` anywhere in the organisation, so intro could not delete its copy while this line stood. This repo's own byte-identical `lectures/graph.txt`, which nothing read, is deleted here too. WHY THIS IS BLOCKED, and what is NOT known. `short_path` cannot be completed in-browser today, and this PR does not change that. `myst-ext-exercise` nests a gated `{exercise-start}` body as children of the `exercise` node — `exercise-start` is an ALIAS for `exercise`, not a hoist — and thebe only wires up cells at the top-level block layer. So the data cell sits at ['root','block','exercise','block'] and never executes, whichever form it takes. On `main` two cells are stranded there; with this change it is one. A second bug currently masks a worse version of that one. `short_path.md:480` is a bare ```{solution}``` closer that mystmd rejects, so the solution node is empty and its five cells spill to root level — which is the ONLY reason they are runnable. Correcting it to the gated pair was tried and rebuilt: the error clears, the cells correctly re-nest, and the page would go from one dead cell to six. So that fix is deliberately NOT included here. RETRACTED, and recorded so it is not repeated: an earlier revision of this PR asserted that `%%file` does not fire under thebe-lite. That was wrong. The evidence was a cell producing no output and a later FileNotFoundError, both of which are equally explained by the cell never having executed — which is what was happening. `%%file` under thebe-lite has never been tested. #65 tests it at top level, where nesting cannot confound the result. Equivalence of the data itself is measured, not assumed: intro's canonical filename-based helper over the file written by this cell gives 286 finite entries, Q[99,99]=0.0 and J[0]=160.5500, identical to the network path this replaces, and the written file is byte-identical at 4,691 bytes, sha256 046759fb... graph.txt is not migrating to QuantEcon/data-lectures: it is synthetic teaching data with null provenance, the exercise prose depends on it being visible, and that repo is a cache for upstream sources rather than a host for our examples.
|
📖 Netlify Preview Ready! Preview URL: https://pr-65--beautiful-dodol-cb9543.netlify.app (1e857e3) ✨ Browse the preview at the URL above. |
The probe page was committed but never reached the toc, so mystmd did not build it and the preview 404s — 42 pages published, none of them the probe. Cause: the edit that was supposed to add it matched on the wrong indentation (` toc:` / ` - file:` where this file has ` toc:` / ` - file:`), and the script printed a success message unconditionally instead of asserting the replacement had happened. A silent no-op reported as done — the same failure mode this probe exists to expose, which is a fair thing to have walked into. This version asserts the anchor matches exactly once and re-reads the file to confirm the entry is present before exiting. Also drops the `.md` extension to match every other toc entry in this file.
|
📖 Netlify Preview Ready! Preview URL: https://pr-65--beautiful-dodol-cb9543.netlify.app (fdd0b9d) ✨ Browse the preview at the URL above. |
The `{note}` container used three backticks around a three-backtick code-cell,
so the inner fence closed the outer directive. Everything after it was absorbed:
the REPORT cell's value began with the literal markdown
## REPORT (top level)
```{code-cell} ipython3
and would have raised SyntaxError rather than reporting anything — the probe
would have failed in a way that says nothing about what it is probing.
A container directive wrapping a code-cell needs MORE backticks than the cell.
The gated `{exercise-start}` / `{solution-start}` forms are unaffected: they are
standalone directives rather than containers, which is why D and E nested
correctly while F's block leaked.
Now asserts its own structure after writing: exactly 7 code cells, none
containing markdown, and the REPORT cell starting where it should.
The page rendered but showed no JupyterLite power button, so nothing on it could
be run — the probe was unrunnable for a reason unrelated to what it probes.
Cause: its frontmatter carried only `title`. Every real lecture here carries
`jupytext` + `kernelspec`, and without a kernelspec mystmd does not treat the
page as a notebook, so thebe never attaches a kernel.
Frontmatter is now copied from short_path.md, the page this probe is modelling.
This is worth recording rather than just fixing, because it is a third distinct
way a cell can fail to execute here, each silent in its own way:
1. no kernelspec -> whole page has no kernel, no run controls
2. nested in a directive -> cell renders with controls but is never wired up
3. cell simply not run -> indistinguishable from a cell that ran and printed
nothing, which is the normal case for an assignment
All three look like "nothing happened". Only the third is the reader's fault,
and the probe's REPORT cell is designed to separate it from the other two.
|
📖 Netlify Preview Ready! Preview URL: https://pr-65--beautiful-dodol-cb9543.netlify.app (6d934fc) ✨ Browse the preview at the URL above. |
mmcky
added a commit
that referenced
this pull request
Aug 12, 2026
…run (#63) * short_path: write the graph with %%file, as every other repo does BLOCKED on #64 — do not merge yet. See that issue and the diagnostic in #65. This PR is the intended END STATE; what is unproven is whether this runtime can run it. This lecture executes in the reader's browser (`lite: true`) and was the only repo in the family fetching this graph across repos on every page view: import requests file_url = "https://raw.githubusercontent.com/QuantEcon/lecture-python-intro/main/lectures/graph.txt" graph_file_response = requests.get(file_url) Every sibling — lecture-python-intro, lecture-dp, lecture-jax, lecture-intro.zh-cn and test-actions-lecture-intro — writes the data with a `%%file graph.txt` cell and reads it back with `open(in_file)`. This repo now does the same, so all six are identical on this exercise, and the divergent string-splitting helper `map_graph_to_distance_matrix(in_file_data)` returns to the canonical filename form. It also retires a cross-repo dependency: that `requests.get` was the only thing reading lecture-python-intro's committed `lectures/graph.txt` anywhere in the organisation, so intro could not delete its copy while this line stood. This repo's own byte-identical `lectures/graph.txt`, which nothing read, is deleted here too. WHY THIS IS BLOCKED, and what is NOT known. `short_path` cannot be completed in-browser today, and this PR does not change that. `myst-ext-exercise` nests a gated `{exercise-start}` body as children of the `exercise` node — `exercise-start` is an ALIAS for `exercise`, not a hoist — and thebe only wires up cells at the top-level block layer. So the data cell sits at ['root','block','exercise','block'] and never executes, whichever form it takes. On `main` two cells are stranded there; with this change it is one. A second bug currently masks a worse version of that one. `short_path.md:480` is a bare ```{solution}``` closer that mystmd rejects, so the solution node is empty and its five cells spill to root level — which is the ONLY reason they are runnable. Correcting it to the gated pair was tried and rebuilt: the error clears, the cells correctly re-nest, and the page would go from one dead cell to six. So that fix is deliberately NOT included here. RETRACTED, and recorded so it is not repeated: an earlier revision of this PR asserted that `%%file` does not fire under thebe-lite. That was wrong. The evidence was a cell producing no output and a later FileNotFoundError, both of which are equally explained by the cell never having executed — which is what was happening. `%%file` under thebe-lite has never been tested. #65 tests it at top level, where nesting cannot confound the result. Equivalence of the data itself is measured, not assumed: intro's canonical filename-based helper over the file written by this cell gives 286 finite entries, Q[99,99]=0.0 and J[0]=160.5500, identical to the network path this replaces, and the written file is byte-identical at 4,691 bytes, sha256 046759fb... graph.txt is not migrating to QuantEcon/data-lectures: it is synthetic teaching data with null provenance, the exercise prose depends on it being visible, and that repo is a cache for upstream sources rather than a host for our examples. * short_path: move the graph data cell out of the exercise so it can run A code cell nested in any directive is never registered with the kernel by @myst-theme/jupyter, so inside `{exercise-start}` this cell renders a run button that does nothing, and the solution then fails with FileNotFoundError. It is the only cell on the page in that position — every other cell is already top level and runs. Moving it below `{exercise-end}` makes the whole lecture runnable in the browser for the first time. Reading order is unchanged: the prose still says "The text below describes a weighted directed graph" and the data still appears directly beneath it, just outside the exercise's box. The solution's "reads in the graph data above" is likewise still accurate. A comment records why it sits there, so nobody tidies it back in — that would silently break the page again, with no error and no CI signal. Root cause and the general fix: #64 and QuantEcon/quantecon-theme.mystmd#117. Verified the move did not touch the data: 9 cells, 100 data lines, reconstitutes to 4,691 bytes / sha256 046759fb..., and the cell now sits strictly between `{exercise-end}` and the solution.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Draft, never for merge. This exists to give #64 a Netlify preview you can click, and to delete afterwards.
Why a probe rather than reasoning
When a cell fails to execute under thebe-lite there is no error and no output — and for an assignment cell, "no output" is also the correct result. Those two states are indistinguishable by eye, which is exactly why this bug took several wrong turns: a
%%filecell producing nothing was read as "the magic is broken" when the cell had simply never run.So this page makes non-execution observable. Every cell appends a marker to a list; a final top-level REPORT cell prints which markers are present and which are
MISSING. One click on that cell names every cell that silently did nothing.What it probes
%%filemagic%%fileactually write to the FS?{exercise-start}short_path{solution-start}short_path— but only because a malformed directive spills it to root{note}How to run it
Open the preview's
probe_nestingpage, run the cells top to bottom, then run the REPORT cell. Expected output shape:What each result would mean
If D is MISSING and A/C/E/F present — nesting inside
{exercise}is the cause, confirming #64, and the fix is to move the data cell out of the exercise block.If D and F are both MISSING — it is nesting generally, not exercises, which is a broader and more interesting upstream report for
mystmd/thebe.If B/C show
%%file worked at top level: True— this is the one that changes the design.%%fileunder thebe-lite has never actually been tested; it was assumed broken on evidence that turned out to be about the cell not running. If it works at top level,lecture-wasmcan converge withlecture-python-intro,lecture-dp,lecture-jax,lecture-intro.zh-cnandtest-actions-lecture-intro— same%%file graph.txtcell, same canonicalopen(in_file)helper — instead of carrying a divergent string-based variant indefinitely.If B/C show
False—%%fileis genuinely unavailable here and the inline literal is the only option, which is worth knowing definitively rather than by assumption.Relates to #63 and #64.