Visualizer: a display of literals has no steps (P40) - #48

What

A list, tuple, set or dict display no longer has a step of its own: once its parts are values, it reads as its value, as on paper. This applies in Ersetzung and Kasten, since both show the same trace.

• A display of literals has no steps, nested at any depth: baum = ["+", ["*", 2, ["-", 7, 3]], ["/", ["+", 1, 2], 3]] has one step, the assignment.

• Negative number literals (-3, -2.5, -1j: unary minus applied to a number literal) have no step, inside a display or anywhere else. ["-", 7, -3] and [-1, 2] have no steps.

• Names, calls and operators keep their steps, and the step that only built the display is gone:

  • [1 + 2, 3, 4] → [3, 3, 4]
  • [a, 2] → [5, 2]
  • [[1, 2], x] → [[1, 2], 7]
  • ["+", 1, ["*", n, 3]] → ["+", 1, ["*", 2, 3]]

  Each is one step.

• The display keeps its step only where its value reads differently from what is written:

  • unpacking: [*b, 2] → [1, 2], {**d, "j": 2};
  • a set or dict that loses duplicates: {1, 1} → {1}.

  These are the only cases where the «build» step changes something.

• Unchanged:

  • dauer[1] → 2;
  • notenschnitt(noten) → notenschnitt([5, 4.5, 6]);
  • the IndexError step for dauer[3].

  [1, 2, 30][1] → 2 is now a single step: the literal's step [1, 2, 30] → [1, 2, 30] changed nothing. The P39 check now expects that.

Fixed along the way (please confirm): unpacking in a display gave the wrong value. [*b, 2] with b = [1] gave [[1], 2]; Python gives [1, 2]. The same happened for tuples and sets. I fixed it because the handlers being rewritten here contained the bug, and without the build step the wrong value would have reached the variables panel unseen. A check compares the result with Python.

Checks

• scripts/check-runtime-lists.py:
  • The baum literal, ["-", 7, -3], [-1, 2], a mixed tuple/dict/set of literals and -3 have no steps except the assignment, and their values are Python's.
  • The four examples above have one step each, plus (n, -n, {n: [n]}) and [len([1, 2]), 3].
  • The unpacking and duplicate cases keep their step, with Python's value.
  • The P39 cases are unchanged except [1, 2, 30][1], which is now one step.
• scripts/check-console-browser.mjs steps through a program with these displays in both Kasten and Ersetzung. Every step must change the code or the variables, and the program has exactly 10 steps (7 assignments, 3 substitutions).
• Locally, lint, test:runtime, test:protocol and test:console pass.

How it goes live

• Courses: merging to main deploys the embed, so the courses get it straight away.
• IDE: set tag v0.1.24 on the merge commit, and I'll open the IDE bump PR.
• Macro: no change.

If P37 (#49) is merged too, one tag covers both.

Manual tests (desktop, and iPad with an external keyboard)

1. Visualize baum = ["+", ["*", 2, ["-", 7, 3]], ["/", ["+", 1, 2], 3]] in Ersetzung, then in Kasten. There is 1 step, and baum appears in the variables.
2. Visualize q = ["-", 7, -3] and r = [-1, 2]. There is one step per line.
3. Visualize the following, stepping with → (keyboard) or the buttons. Each line shows one substitution, then the assignment, and every step changes something:

   x = [1 + 2, 3, 4]
   a = 5
   y = [a, 2]
   z = [[1, 2], a]
   n = 2
   w = ["+", 1, ["*", n, 3]]

4. P39 is unchanged:
   • dauer = [1, 2, 30], x = dauer[1] goes dauer[1] → 2;
   • [1, 2, 30][1] is one step to 2;
   • dauer[3] ends at the IndexError step;
   • notenschnitt(noten) → notenschnitt([5, 4.5, 6]).
5. Options → «Kopieren → Kette» for item 3: the chain has the lines shown, with no line repeated.

Follow-ups (not fixed)

• -x with x = 7 still has a step that changes nothing. After x → 7 the line reads -7, and the step -7 → -7 follows. I left it out because skipping it whenever the text matches would leave -(a + b) showing -(3). To decide: drop the unary-minus step when its operand is a name.
• +3 (unary plus) keeps its step +3 → 3, which does change the text. Say if it should count as a literal too.
• CI, first run: test:console failed in the input() check from P36, which this PR doesn't touch, and passed on a rerun. The console's text included the entry's «Visualisieren» button label before the input lines (">>> repl()\nVisualisieren\n> 1 + 2…"). That check is timing-dependent; it should read the transcript without the buttons.

🤖 Generated with Claude Code