UPSTREAM PR #17558: abstract env var names so they're insensitive on windows and sensitive on other operating systems

[loci-dev]

Note

Source pull request: nushell/nushell#17558

Apparently, there are still edge cases for case-insensitive env vars in Windows, so someone pointed me a different way to try and do it. This PR represents that way. All env vars (not just path) will be treated case-insensitive on Windows and case-sensitive on all other operating systems. Since this does the sensitivity checking for us, I removed (commented out for now) the get_env_var_insensitive function. I realize that windows is case-preserving and not really case-insensitive but this seemed like the best option for now.

Let me know if this is a bad idea.

Release notes summary - What our users need to know

Tasks after submitting

• Update the documentation

[loci-review]

Overview

Analysis of 39,976 functions in Nushell (3,004 modified, 5,224 new, 5,218 removed, 26,530 unchanged) reveals mixed performance impact from environment variable refactoring across three commits implementing case-insensitive handling on Windows. Binary target.aarch64-unknown-linux-gnu.release.nu shows +0.248% power consumption (+2,294 nJ). Changes introduce EnvName wrapper type replacing O(n) linear scans with O(1) HashMap lookups, achieving correctness improvements with acceptable performance trade-offs.

Function Analysis

Major Regressions (Job Control System):

• Mailbox::clear(): Response time 4,128ns → 26,628ns (+545%, +22,500ns), throughput stable at 69ns. Regression in child functions (BTreeMap/HashMap clear operations) due to PipelineData complexity with EnvName type.
• Mailbox::recv_timeout(): Response time 11,208ns → 54,010ns (+382%, +42,802ns), throughput stable at 273ns. MPSC channel operations affected by PipelineData handling changes.
• Mailbox::try_recv(): Response time 11,462ns → 54,260ns (+373%, +42,798ns), throughput stable at 108ns. Similar pattern to recv_timeout with unchanged source code.

Major Improvements (Command System):

• Alias::run(): Response time 24,963ns → 417ns (-98%, -24,546ns), throughput stable at 94ns. Optimized error propagation benefits guard function despite no source changes.
• Predeclaration::run(): Response time 24,891ns → 2,393ns (-90%, -22,498ns). Reduced environment variable lookup overhead in call hierarchy.
• BlockCommand::run(): Response time 24,918ns → 2,422ns (-90%, -22,496ns). Faster error handling infrastructure.

Hot-Path Optimization:

• type_compatible(): Response time 856ns → 297ns (-65%, -559ns), throughput 207ns → 101ns (-51%, -106ns). Performance-critical function in IR execution pipeline shows compiler optimization success.

Direct Optimization:

• get_editor(): Response time 24,544ns → 22,450ns (-9%, -2,094ns), throughput 530ns → 245ns (-54%, -285ns). Refactored from get_env_vars() to targeted get_env_var() calls, eliminating unnecessary HashMap allocation.

Other analyzed functions in error handling, type comparison, and compiler-generated code showed minor changes with limited practical impact.

Additional Findings

Job control regressions (373-545%) occur in non-critical paths outside primary hot paths (IR execution, pipeline flow, parser, REPL loop). The 42-43µs absolute increases are acceptable for cleanup operations. Error handling shows net positive impact: guard functions improve 90-98% (executing frequently) while diagnostic functions regress 128-176% (executing rarely). The architectural refactoring successfully optimizes primary use cases while accepting minor secondary system regressions, justified by cross-platform correctness requirements.

🔎 Full breakdown: Loci Inspector.
💬 Questions? Tag @loci-dev.