parser: a case pattern may be produced by a substitution (#744)

[berrym]

Closes #744.

The defect

A case pattern is a word, so a substitution is a legal part of one. None of the three spellings parsed:

$ lush -c 'case a in $(printf a)) echo match;; esac'
error[E1001]: expected pattern in case statement

Same for the backtick form and $((1+2)). bash, zsh and dash accept all three.

Cause

The case-pattern parser carries a hand-written list of accepted token types. It included TOK_VARIABLE — which is why $var and ${p} worked, and what made the gap look narrower than it was — but none of TOK_COMMAND_SUB, TOK_BACKQUOTE, TOK_ARITH_EXP.

The two canonical predicates next door in tokenizer.c, token_is_argument_word_token and token_is_assignment_value_token, both already list all three. Only this site missed them. Added here rather than by adopting a predicate, because this list is genuinely its own: it also accepts *, [[, ]] and excludes keywords.

Not the #494 shape, and the issue guessed wrong

I filed this pointing at the pattern scanner, reasoning it failed to treat a backtick body as opaque. The scanner was never involved. #494's )-boundary work is intact — once the tokens are accepted, a ) inside a substitution correctly does not end the pattern:

$ lush -c 'case ")" in $(printf ")")) echo match;; esac'   -> match
$ lush -c 'case ")" in `printf ")"`) echo match;; esac'    -> match

Both are pinned in the test so the two defects cannot be confused again.

The issue also claimed the $( ) spelling "parses and matches" — a control I asserted without running. Every substitution form failed identically. Scope corrected on the issue.

Verification

• tests/integration/test_case_pattern_substitution.c — 21 checks: all three spellings as a whole pattern and as part of one, alternation, nesting, a non-matching pattern that must still fall through (so the fix is not "accept and always match"), glob metacharacters on the produced pattern, the two Command substitution mis-terminates at a case-pattern ) #494 boundary cases, and the eight forms that already worked.
• 13 of 21 fail against the parent build, with a control proving the binaries differ.
• Full suite 195/195. ASan 65/65; the new test 1/1 under ASan.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!