[sf624]

Refactor step 3.2, a fourth layer of the stack (#240, #241, #242): two findings of the reviews of 3.2.

processOf is kept per environment and per call link (src/semantics/env.ts). It read the chains of every call of a function again for each of its callers' calls. In a chain of functions where each one runs the one before in $( ) and calls the one before that plainly (f3() { local a=$(f2); f1; }), that doubled at each function: 40 of them took 52–54 s in npx tsx test/dump.ts (16 s with if x=$(…)). Now 0.7–0.9 s for every shape tried (plain, set -e, if, ||, &&, |, ( ) &). Environments and call links never change once made, so the results are the same: every probe test passes, and the dumps of the fixtures and of 21 chain scripts (12, 20, 25 functions, each shape; 40 for three of them) are identical before and after.
Test: test/env.test.ts "explains a long chain of functions, each called in $( ) and plainly, in little time" (each shape, 40 functions, every line, under 5 s; it took 85 s before).

The timing test of the arithmetic bound is measured against a baseline of the same run (test/state.test.ts "gives up on an expression too long or too deep to follow, quickly"). It failed on loaded machines (3.5 s against 3 s; 1.3 s alone). Now 600 lines past the bound must take less than three times as long as 60 lines (followed, they would take ten times as long). A failing test also skipped the cleaning of the coverage report, which is why src/wordfacts.ts:40 looked uncovered then; it is covered when the tests pass.

The same for test/state.test.ts "is fast on long scripts with many functions" (300 functions against 150) and test/nested.test.ts "takes a time linear in the strings…" (2,000 lines against 200), which also failed under load.

CHANGELOG: one Fixed line (the freeze on such scripts).

🤖 Generated with Claude Code

[github-actions]

Web preview: https://sf624.github.io/bash-explainer/pr/246/ (commit c24a9e3)
Compare with main: https://sf624.github.io/bash-explainer/pr/246/compare.html