[rocky]

Resolve the merge conflict by accepting new tests that have been added to master.
Perform some Ruff and mypy fixes that are not in a merge to a follow-on PR that has mathics.core.list ruff and mypy changes.

[rocky]

Merged into branch directly.

[tombouctou]

I pushed two commits to this branch after a review. Each commit message has the details.

ecbbd9e28: merge of master

master already has #2084, which makes the same set_element/_copy change, so I resolved the conflicts in mathics/core/list.py and mathics/core/subexpression.py by taking master's version. That removes:

• _is_atom and _part_positions defined twice (the second pair silently replaced the first; Ruff F811);
• two mypy errors in list.py: list(self.value) without the or () guard, and element.value on a BaseElement;
• the Optional/Tuple imports and # -*- coding: utf-8 -*- headers that master had removed.

The # noqa: ISC004 in test_eol.py is kept.

a264d2ba2: spans checked as in WMA; a shared path is copied once

Spans in Part assignment (also wrong on master):

• A span reaching before the start of the list replaced the head: pa = Range[5]; pa[[-10;;-1]] = 0 gave 0[0, 0, 0, 0, 0], and pa[[-6;;2]] = x gave x[x, x, 3, 4, 5]. A span outside the list now gives Part::take and leaves the value unchanged.
• An end that is left out or given as All is -1 whatever the step, so pa[[;; ;; -1]] = ... and pa[[3;;All;;-1]] = 0 now give take instead of silently doing nothing.
• A span may select nothing only right next to the elements: 3;;2, 6;;5, 1;;-6 and 5;;6;;-1 are allowed; 7;;6, -7;;-7, 0;;2 and 4;;2 are errors.
• About 30 cases were checked against WMA 15 (wolframscript), and most of them are now tests in test_eol.py. One existing test changed: pa[[2 ;; 5]] = 0 on a list of 3 now expects WMA's take message instead of partw.
• WMA reports these as Set::take, but here they come out as Part::take, the same way partw already does. That is left for a separate change.

Copying:

• An expression on the way to several parts was copied again for each part. One assignment now remembers the copies it made, by id, keeping the objects alive so an id cannot be reused. m[[1, All]] = 1 on a 300×2000 matrix went from 0.17 s to 0.07 s. A test checks that the shared original is left untouched.
• _copy of a List now calls shallow_copy() instead of repeating its code. Other expressions still use Expression(head, *elements): Expression.set_element drops the copy's properties and cache straight away, so shallow_copy, which rebuilds the cache, would only add work.
• The new annotations make mypy check replace, so _headp/_elementsp are now declared and current_index is narrowed.

Left as is on purpose: ListExpression.set_element keeps the old element's symbols in the cache, as decided in #2084. Rescanning the list on every assignment would cost more than the occasional extra re-evaluation.

The list, assignment, core and evaluation tests pass locally (709 passed, 6 xfailed). The full pytest and doctest run is still going.

🤖 Generated with Claude Code