[eliotwrobson]

A correct plot that also has plt.text(...) is currently rejected when it's sent to the test, so the student scores 0 with "text and annotations aren't supported". A review comment on PrairieLearn/PrairieLearn#16284 found this. axhline and axvline, which students often use for reference lines, are rejected the same way.

• Sent to the test:
  • text in data or axes coordinates, with its alignment and rotation.
  • annotate, with its point in data or axes coordinates, its text position in data or axes coordinates or offset points, and whether it has an arrow. Every arrow comes back as a plain -> arrow.
  • axhline and axvline, with their position, extent, label, color, width and style. A line is treated as one only if it uses their blended transform and has a constant value, so other blended lines are still rejected.
  • Text in figure coordinates, axline, errorbar and fill_between are still rejected.
• assert_plot_equal:
  • axhline and axvline lines are compared as their own kinds, with the same rules as other lines.
  • The new check_texts option, off by default, compares the strings of text and annotations, and the point each annotation points at. The position of plain text isn't compared.
  • Extra text never fails a comparison unless check_texts is on, which covers the review comment's case.
• Startup timeout race: the start response was read with the same timeout the sandbox used for the student code, so a slow submission got either "Student code initialization timed out" or "No response from initialization" at random. This made test_timeout flaky. The read now waits an extra 0.5 seconds, so the sandbox's own message is reported. "No response" is now only for a sandbox that doesn't answer at all.

Testing

With an image built from this branch, the converted plots question gives:

Submission	Score
Starter code plus plt.text	1.0
Starter code plus annotate with an arrow	1.0
Starter code plus axhline	0.0: Expected 0 horizontal lines in your plot, but found 1.

The axhline result is intended: extra lines fail unless the test sets allow_extra. In all three cases, the image shown with the results matched the student's plot.

The timeout scenarios failed about 1 run in 12 before the race fix, and passed 40 runs in a row after it.

🤖 Generated with Claude Code