Add ${var op word} alphabet matrix (mix bash_fixtures.gen varop)

[davydog187]

Implements #70 item 2 for ${var op word} parameter expansion only: a generated varop_matrix suite in the same generate / record / test path as date_matrix, printf_matrix, and test_matrix.

This is not the item-3 filesystem-shape × command cube, Oils activation, or CI topology.

What was enumerated

mix bash_fixtures.gen varop writes 270 cases. Suite name is varop_matrix (not param_matrix, to keep it distinct from the hand-written parameter_expansion suite).

• Word-taking operators: - :- = := + :+ ? :? × unset / set-empty / foo / foobar × nonempty vs empty word
• ${v:-1} next to ${v: -1} so a parser cannot hide the colon/space distinction behind word
• Length ${#v} at the same four states
• Substring ${v:offset} / ${v:offset:len} including :3 (empty vs bar), :1:3, : -N, ::2, :0:-1
• Prefix/suffix # ## % %% × literal / glob / empty pattern, plus foofoo where shortest vs longest actually splits
• Replacement / // × first-vs-all (o occurs twice), match vs miss, delete, empty-pattern forms, and foofoo for first vs all of foo
• set -u cells so unset-safe word-ops stay recorded next to ${#v} / substring / removal / replacement, which bash still treats as unbound
• Every expansion is double-quoted (echo "[${v-word}]"); set +u is explicit except the nounset cells

mix bash_fixtures.gen varop --dry-run reports varop_matrix: 270 cases. mix bash_fixtures.gen now generates date, printf, test, and varop.

Recording

Docker was not available on the agent VM. Oracle output was recorded with the same test/fixtures/runner.sh the Docker task runs, under GNU bash 5.2.21 / Ubuntu 24.04 (uid ubuntu, not root), then pretty-printed through Mix.Tasks.BashFixtures.write_json!/2:

mix bash_fixtures.gen varop
bash test/fixtures/runner.sh < test/fixtures/bash_cases/varop_matrix.json > /tmp/varop_matrix_raw.json
# then Jason-pretty-print + Fixtures.validate, as `mix bash_fixtures` does

To re-record through Docker when it is available:

mix bash_fixtures varop_matrix

Known gaps (18 of 270)

No cell was omitted. Divergences are opts.known_gap (excluded from the digest). The fixture runner still executes them with the assertion inverted.

Count	Reason
9	${v?} / ${v:?} returns empty and continues; bash writes a diagnostic and aborts
4	set -u on ${#v} / ${v:0} / ${v#x} / ${v/x/y} of unset: JustBash treats empty; bash errors unbound variable
4	Shortest #* / %* consumes one character; bash's shortest * match is empty so the value is unchanged
1	${v:0:-1} on empty is empty at exit 0; bash errors substring expression < 0

Matching siblings are left unmarked (for example ${v?} on a set value, ${v##*}, ${v:0:-1} on foo).

JustBash.exec/2 did not raise on any cell. No expansion implementation bugs were fixed in this PR — it is the enumeration + harness.

Cheap follow-ups the matrix now makes un-skippable

• ${v?} / ${v:?} should write the diagnostic and abort
• set -u should error on ${#unset} / ${unset:offset} / ${unset#pat} / ${unset/pat}
• Shortest #* / %* should match the empty string
• Negative substring length on an empty value should be a diagnostic, not silent empty

Not in this PR: the item-3 FS-shape × command cube, Oils activation, CI topology changes.

Tests

All local quality gates passed:

• mix compile --warnings-as-errors
• mix format --check-formatted
• mix credo --strict (only the existing intentional banned_fixture_apply finding)
• mix dialyzer (zero new errors)
• mix test — 6622 tests, 0 failures
• mix bash_fixtures.gen varop --dry-run — varop_matrix: 270 cases

Related Issues

Related to #70 (item 2, ${var op word} only; not closing the issue)

Type of Change

• New feature (non-breaking change which adds functionality)

Testing

• Added new tests
• All existing tests pass
• Tested manually (mix bash_fixtures.gen varop --dry-run, local runner.sh recording, JustBash vs oracle classification)

Checklist

• My code follows the style guidelines of this project
• I have run mix format
• I have run mix credo and addressed any issues
• I have added tests that prove my fix is effective or that my feature works
• New and existing tests pass locally with my changes
• I have updated the documentation accordingly
• I have updated the CHANGELOG.md (for non-trivial changes)