fix(variables): keep dynamic well-known vars live across declare -a / array assign

[lu-zero]

Problem

Once a script declare -as or plain-array-assigns a dynamic well-known variable such as PIPESTATUS, brush never updates it again on later pipelines in that shell — it stays frozen at whatever was last assigned. Real bash keeps the variable live and refreshes it on every pipeline.

Minimal repro:

declare -a PIPESTATUS=([0]="1")
true | true | true
echo "${PIPESTATUS[@]}"

• bash: 0 0 0
• brush (current origin/main): 1 (frozen)

Same freeze happens with a plain array assignment (no declare):

PIPESTATUS=([0]="9")
true | false | true
echo "${PIPESTATUS[@]}"

• bash: 0 1 0
• brush: 9

Root cause

PIPESTATUS (and other dynamic well-known variables: RANDOM, SECONDS, FUNCNAME, BASH_LINENO, BASH_SOURCE, BASH_ALIASES, …) are backed by a ShellValue::Dynamic { getter, setter } whose getter is consulted on every read and whose setter is a no-op (|_| ()) for all of them.

Two write paths in brush-core/src/variables.rs unconditionally overwrote self.value with a static array, discarding the Dynamic binding and permanently freezing the variable:

1. convert_to_indexed_array — the _ arm hit for Dynamic values and replaced self.value. Triggered by declare -a VAR=(...) via brush-builtins/src/declare.rs.
2. assign (non-append Array-literal arm) — explicitly matched ShellValue::Dynamic { .. } in the overwrite set. Triggered by a plain VAR=([0]="...") assignment.

From that point on the getter closure was gone, so the variable stopped refreshing for the rest of the shell's life.

Fix

• convert_to_indexed_array: add an explicit ShellValue::Dynamic { .. } => Ok(()) arm so the binding survives.
• assign: drop ShellValue::Dynamic { .. } from the overwrite set; the existing (ShellValue::Dynamic { .. }, _) => Ok(()) arm below now handles it correctly.

convert_to_associative_array intentionally unchanged

convert_to_associative_array has the same overwrite pattern, but real bash is asymmetric here: declare -A PIPESTATUS=([x]="9") does permanently freeze it in bash itself (PIPESTATUS becomes a plain associative array that pipelines don't refresh). Pre-fix brush already matched bash's -A behavior, so changing it would diverge from bash compat.

Verification

All four behavioral cases now match bash:

op	bash	brush (after)
declare -a then pipe	0 0 0	0 0 0 ✓
declare -A then pipe	9	9 ✓
plain =([0]="9") then pipe	0 1 0	0 1 0 ✓
plain pipe	0 0 0	0 0 0 ✓

Adds three compat test cases under brush-shell/tests/cases/compat/well_known_vars.yaml (declare -a, plain array assign, declare -A). cargo fmt --check clean.

This surfaced in the wild: a worker-env dump/restore mechanism restored a declare -a PIPESTATUS=([0]="1") line and a downstream pipestatus || die check fired incorrectly, misreporting a successful pipeline as failed.

[lu-zero]

The Source code checks (1.88.0, ubuntu) failure is cargo deny --all-features check all flagging two newly-published advisories (RUSTSEC-2026-0194, RUSTSEC-2026-0195) in the advisory DB. bans, licenses, and sources all pass. This is unrelated to this change (a 2-line variables.rs edit + tests) and would fail on origin/main too — it is an upstream advisory-DB/infra issue affecting all open PRs.

[github-actions]

Performance Benchmark Report

Benchmark name	Baseline (μs)	Test/PR (μs)	Delta (μs)	Delta %
clone_shell_object	17.40 μs	17.25 μs	-0.15 μs	⚪ Unchanged
eval_arithmetic	0.15 μs	0.15 μs	0.00 μs	⚪ Unchanged
expand_one_string	1.80 μs	1.76 μs	-0.04 μs	⚪ Unchanged
for_loop	30.59 μs	30.84 μs	0.25 μs	⚪ Unchanged
full_peg_complex	57.51 μs	58.00 μs	0.49 μs	🟠 +0.85%
full_peg_for_loop	6.20 μs	6.20 μs	0.00 μs	⚪ Unchanged
full_peg_nested_expansions	16.25 μs	16.06 μs	-0.19 μs	🟢 -1.16%
full_peg_pipeline	4.25 μs	4.25 μs	0.00 μs	⚪ Unchanged
full_peg_simple	1.77 μs	1.78 μs	0.01 μs	⚪ Unchanged
function_call	3.82 μs	3.61 μs	-0.21 μs	⚪ Unchanged
instantiate_shell	54.86 μs	54.37 μs	-0.49 μs	🟢 -0.90%
instantiate_shell_with_init_scripts	27352.41 μs	26770.31 μs	-582.11 μs	⚪ Unchanged
parse_peg_bash_completion	2108.11 μs	2112.12 μs	4.01 μs	⚪ Unchanged
parse_peg_complex	20.52 μs	20.68 μs	0.16 μs	🟠 +0.76%
parse_peg_for_loop	2.01 μs	2.03 μs	0.02 μs	🟠 +1.19%
parse_peg_pipeline	2.15 μs	2.13 μs	-0.02 μs	⚪ Unchanged
parse_peg_simple	1.09 μs	1.09 μs	-0.00 μs	⚪ Unchanged
run_echo_builtin_command	16.51 μs	16.16 μs	-0.35 μs	⚪ Unchanged
tokenize_sample_script	3.37 μs	3.41 μs	0.04 μs	⚪ Unchanged

Code Coverage Report: Only Changed Files listed

Package	Base Coverage	New Coverage	Difference
brush-core/src/variables.rs	🟢 91.38%	🟢 91.57%	🟢 0.19%
Overall Coverage	🟢 74.81%	🟢 74.81%	⚪ 0%

Minimum allowed coverage is 70%, this run produced 74.81%
Maximum allowed coverage difference is -5%, this run produced 0%

Test Summary: bash-completion test suite

Outcome	Count	Percentage
✅ Pass	1583	75.06
❗️ Error	17	0.81
❌ Fail	155	7.35
⏩ Skip	339	16.07
❎ Expected Fail	13	0.62
✔️ Unexpected Pass	2	0.09
📊 Total	2109	100.00

[Copilot]

Pull request overview

Fixes PIPESTATUS becoming static after indexed-array declarations or assignments.

Changes:

• Preserves dynamic values during indexed-array conversion and assignment.
• Adds compatibility tests for indexed, associative, and plain assignments.
• Review found the conversion is incorrectly applied to all dynamic variable types.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
brush-core/src/variables.rs	Adjusts dynamic variable conversion and assignment.
brush-shell/tests/cases/compat/well_known_vars.yaml	Adds PIPESTATUS regression tests.