[tombouctou]

Evaluation.evaluate(query, timeout=...) (and MathicsSession.evaluate_as_in_cli(..., timeout=...)) never used its timeout; nothing stopped the evaluation when it ran out. A caller that wanted a limit had to interrupt from outside, e.g. with SIGALRM, at an arbitrary point in the Python code. An interrupt in the middle of an iteration step could leave Table's step counter as an unevaluated Plus[...], which then crashed eval_iter with TypeError: '>' not supported between instances of 'NoneType' and 'int'. I hit this while running the examples of Computational Discrete Mathematics with Combinatorica, with a per-cell time limit.

• The timeout is now a deadline that check_stopped checks, between evaluation steps. The evaluation ends with $Aborted and General::timeout, and the deadline is cleared, so the next evaluation has no limit left over.
• The iterator's step counter is now a Python int. This also saves evaluating Plus[index, 1] on every step.

settings.TIMEOUT is None by default, so the CLI and REPL behave as before.

New test: Do[k^2, {k, 10^7}] and a Table of Timings with timeout=0.3 give $Aborted and "Timeout reached." in 0.3 s, and the next evaluation runs without a limit.

🤖 Generated with Claude Code

[mmatera]

@tombouctou, thanks for checking this out. Currently, timeout is implemented using stopit. The function doing this task is in

mathics-core/mathics/eval/datetime.py

Line 68 in 7286b06

if sys.platform == "emscripten":

At some point we move the call into the code of TimeConstrained, and forgot to handle the timeout parameter in the evaluation.

[rocky]

LGTM. And thanks!