Doctest # optional / # long annotation dataflow checker

[mkoeppe]

Is there an existing issue for this?

• I have searched the existing issues for a bug report that matches the one I want to file, without success.

Problem Description

It is easy to make mistakes when using # optional - FEATURE and # long time annotations in doctests.

For example, in the following doctests, a # long time annotation is missing, causing a failure when sage -t is used without --long.

src/sage/coding/ag_code_decoders.pyx:    sage: code = codes.EvaluationAGCode(pls, G)  # long time
src/sage/coding/ag_code_decoders.pyx:    sage: code                                   # long time
src/sage/coding/ag_code_decoders.pyx-    [26, 15] evaluation AG code over GF(9)
src/sage/coding/ag_code_decoders.pyx:    sage: decoder = code.decoder('K')
src/sage/coding/ag_code_decoders.pyx:    sage: tau = decoder.decoding_radius()        # long time
src/sage/coding/ag_code_decoders.pyx:    sage: tau                                    # long time

As part of

• Meta-ticket: Modularize sagelib into separate distributions (pip-installable packages) sagemath-... for Sage 10.x #29705

we expect to add many more # optional annotations corresponding to "standard" packages such as # optional - scipy and corresponding to subset distributions such as # optional - sage.libs.pari.

Proposed Solution

The Sage doctester has a rudimentary data flow analyzer for the globals involved in the lines of a doctest, globs: RecordingDict, which is used for determining dependencies for parallelization purposes.

We can extend this mechanism to track the set of features (# optional tags) and the pseudo-feature long associated with each doctest line. For each variable in RecordingDict.got of a doctest_line, we check that the set of required features

• is a superset of the set of required features of a predecessor of doctest_line that has variable in predecessor.globs.set,
• or a proper generalization of this.

Alternatives Considered

N/A

Additional Information

No response