perf: experiment with immutable strings

[reubeno]

No description provided.

[github-actions]

Test Results

    2 files      9 suites   1m 17s ⏱️
  569 tests   569 ✅ 0 💤 0 ❌
1 124 runs  1 124 ✅ 0 💤 0 ❌

Results for commit 03d57c7.

[github-actions]

Performance Benchmark Report

Benchmark name	Baseline (μs)	Test/PR (μs)	Delta (μs)	Delta %
expand_one_string	3.52 μs	3.51 μs	-0.00 μs	⚪ Unchanged
for_loop	587665.96 μs	578031.33 μs	-9634.63 μs	🟢 -1.64%
instantiate_shell	62.10 μs	62.03 μs	-0.07 μs	⚪ Unchanged
instantiate_shell_with_init_scripts	28430.04 μs	29777.49 μs	1347.45 μs	🟠 +4.74%
parse_bash_completion	2806.87 μs	2661.94 μs	-144.93 μs	🟢 -5.16%
parse_sample_script	4.26 μs	4.42 μs	0.16 μs	🟠 +3.68%
run_echo_builtin_command	92.11 μs	92.53 μs	0.43 μs	⚪ Unchanged
run_one_builtin_command	113.77 μs	110.22 μs	-3.54 μs	⚪ Unchanged
run_one_external_command	1754.84 μs	1815.26 μs	60.41 μs	🟠 +3.44%
run_one_external_command_directly	826.52 μs	825.19 μs	-1.33 μs	⚪ Unchanged

Code Coverage Report: Only Changed Files listed

Package	Base Coverage	New Coverage	Difference
brush-core/src/builtins/declare.rs	🟢 82.63%	🟢 82.83%	🟢 0.2%
brush-core/src/builtins/export.rs	🟠 74.07%	🟢 75.44%	🟢 1.37%
brush-core/src/builtins/unset.rs	🟢 85.11%	🟢 85.71%	🟢 0.6%
brush-core/src/completion.rs	🟠 63.1%	🟠 63.21%	🟢 0.11%
brush-core/src/expansion.rs	🟢 96.1%	🟢 96.13%	🟢 0.03%
brush-core/src/interp.rs	🟢 92.35%	🟢 92.36%	🟢 0.01%
brush-core/src/prompt.rs	🟢 90.38%	🟢 90.26%	🔴 -0.12%
brush-parser/src/ast.rs	🔴 25.35%	🔴 24.91%	🔴 -0.44%
brush-parser/src/tokenizer.rs	🟢 96.12%	🟢 96.28%	🟢 0.16%
Overall Coverage	🟢 77.34%	🟢 77.35%	🟢 0.01%

Minimum allowed coverage is 70%, this run produced 77.35%

[reubeno]

@39555 I'm interested to get your take since you've been looking at and thinking about parsing, tokenizing, and strings. Independent of parser approach (I.e. winnow/nom, peg, etc.), what do you think about looking at one of the many "low-cost clone" / CoW string crates? I encountered imstr but also see arcstr and a variety of others in similar category.

Given how much of the shell is string manipulation, I could see benefit but am concerned about how virally they spread in a code base and how much we'd get locked into them.

This draft is just an experiment in the parser crate. I imagine we'd see greater benefit by going more widespread and figuring out how to better work with them, where we can use the non-thread-safe variants, etc.