Work around $PATH issues under WSL

[mqudsi]

A common complaint has been the massive amount of directories Windows appends to $PATH slowing down fish when it attempts to find a non-existent binary (which it does a lot more often than someone not in the know might think). The typical workaround suggested is to trim unneeded entries from $PATH, but this a) has considerable friction, b) breaks resolution of Windows binaries (you can no longer use clip.exe, cmd.exe, etc).

This patch introduces a two-PATH workaround. If the cmd we are executing does not contain a period (i.e. has no extension) it by definition cannot be a Windows executable. In this case, we skip searching for it in any of the auto-mounted, auto-PATH-appended directories like /mnt/c/Windows/ or /mnt/c/Program Files, but we do include those directories if what we're searching for could be a Windows executable. (For now, instead of hard-coding a list of known Windows executable extensions like .bat, .cmd, .exe, etc, we just depend on the presence of an extension at all).

e.g. this is what starting up fish prints with logging enabled (that has been removed):

bypassing 100 dirs for lookup of kill
bypassing 100 dirs for lookup of zoxide
bypassing 100 dirs for lookup of zoxide
bypassing 100 dirs for lookup of fd
not bypassing dirs for lookup of open.exe
not bypassing dirs for lookup of git.exe

This has resulted in a massive speedup of common fish functions, especially anywhere we internally use or perform the equivalent of if command -q foo and there's a decent chance foo doesn't exist.

Note that the is_windows_subsystem_for_linux() check will need to be patched to extend this workaround to WSLv2, but I'll do that separately.

Note also that I've taken approach that tries to exempt Windows-mounted /mnt/x/... paths manually added to $fish_user_paths from being excluded, just in case someone downloaded Linux binaries to /mnt/c/Users/jsmith/Downloads/mingw-something-something-toolchain-x.y.z/bin/ directory and wants to use them from Linux without a relative or absolute path. Relative and absolute paths continue to resolve directly, so the surface area of this change is actually a lot smaller than it would seem.

Benchmarks to follow.

Closes #10503

[mqudsi]

--- bench-post  2024-05-18 15:07:05.260920400 -0500
+++ bench-post2 2024-05-18 15:35:07.510279200 -0500
@@ -1,70 +1,70 @@
 aliases.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/aliases.fish
-  Time (mean ± σ):     413.1 ms ±   2.7 ms    [User: 256.1 ms, System: 200.2 ms]
-  Range (min … max):   409.6 ms … 419.3 ms    10 runs
+  Time (mean ± σ):     413.5 ms ±   2.1 ms    [User: 226.5 ms, System: 224.1 ms]
+  Range (min … max):   410.9 ms … 417.9 ms    10 runs

 external_cmds.fish
 Benchmark #1: release/fish --no-config 'benchmarks/benchmarks/external_cmds.fish'
-  Time (mean ± σ):      2.741 s ±  0.015 s    [User: 135.8 ms, System: 2563.2 ms]
-  Range (min … max):    2.719 s …  2.762 s    10 runs
+  Time (mean ± σ):      2.485 s ±  0.021 s    [User: 109.2 ms, System: 2304.1 ms]
+  Range (min … max):    2.455 s …  2.518 s    10 runs

 glob.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/glob.fish
-  Time (mean ± σ):      1.208 s ±  0.006 s    [User: 228.0 ms, System: 979.8 ms]
-  Range (min … max):    1.199 s …  1.219 s    10 runs
+  Time (mean ± σ):      1.233 s ±  0.037 s    [User: 212.4 ms, System: 1019.5 ms]
+  Range (min … max):    1.204 s …  1.330 s    10 runs

 load_completions.fish
 Benchmark #1: release/fish --no-config 'benchmarks/benchmarks/load_completions.fish'
-  Time (mean ± σ):     15.825 s ±  0.131 s    [User: 1.331 s, System: 14.513 s]
-  Range (min … max):   15.676 s … 16.100 s    10 runs
+  Time (mean ± σ):      3.691 s ±  0.006 s    [User: 1.191 s, System: 2.494 s]
+  Range (min … max):    3.682 s …  3.704 s    10 runs

 math.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/math.fish
-  Time (mean ± σ):      1.045 s ±  0.014 s    [User: 754.6 ms, System: 291.0 ms]
-  Range (min … max):    1.012 s …  1.061 s    10 runs
+  Time (mean ± σ):      1.041 s ±  0.018 s    [User: 726.5 ms, System: 313.4 ms]
+  Range (min … max):    1.006 s …  1.063 s    10 runs

 no_execute.fish
 Benchmark #1: release/fish --no-config 'benchmarks/benchmarks/no_execute.fish'
-  Time (mean ± σ):      1.022 s ±  0.019 s    [User: 314.0 ms, System: 671.1 ms]
-  Range (min … max):    0.999 s …  1.057 s    10 runs
+  Time (mean ± σ):      1.003 s ±  0.002 s    [User: 299.9 ms, System: 658.5 ms]
+  Range (min … max):    0.999 s …  1.007 s    10 runs

 printf-escapes.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/printf-escapes.fish
-  Time (mean ± σ):      26.2 ms ±   0.4 ms    [User: 9.7 ms, System: 16.3 ms]
-  Range (min … max):    25.7 ms …  27.6 ms    103 runs
+  Time (mean ± σ):      26.2 ms ±   0.3 ms    [User: 8.8 ms, System: 17.1 ms]
+  Range (min … max):    25.6 ms …  27.8 ms    101 runs

 printf.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/printf.fish
-  Time (mean ± σ):     960.5 ms ±  14.1 ms    [User: 676.4 ms, System: 288.0 ms]
-  Range (min … max):   943.3 ms … 992.8 ms    10 runs
+  Time (mean ± σ):     966.0 ms ±  11.2 ms    [User: 657.6 ms, System: 308.7 ms]
+  Range (min … max):   951.5 ms … 988.8 ms    10 runs

 read.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/read.fish
-  Time (mean ± σ):      3.322 s ±  0.071 s    [User: 165.6 ms, System: 3063.2 ms]
-  Range (min … max):    3.232 s …  3.417 s    10 runs
+  Time (mean ± σ):      3.259 s ±  0.012 s    [User: 171.8 ms, System: 2977.3 ms]
+  Range (min … max):    3.247 s …  3.286 s    10 runs

 seq_echo.fish
 Benchmark #1: release/fish --no-config 'benchmarks/benchmarks/seq_echo.fish'
-  Time (mean ± σ):      89.1 ms ±   1.1 ms    [User: 41.7 ms, System: 46.0 ms]
-  Range (min … max):    87.5 ms …  91.9 ms    32 runs
+  Time (mean ± σ):      89.4 ms ±   1.2 ms    [User: 46.8 ms, System: 42.0 ms]
+  Range (min … max):    87.6 ms …  92.7 ms    32 runs

 set_long.fish
 Benchmark #1: release/fish --no-config 'benchmarks/benchmarks/set_long.fish'
-  Time (mean ± σ):     405.8 ms ±   2.9 ms    [User: 349.9 ms, System: 53.6 ms]
-  Range (min … max):   402.9 ms … 410.8 ms    10 runs
+  Time (mean ± σ):     404.4 ms ±   3.0 ms    [User: 359.4 ms, System: 44.1 ms]
+  Range (min … max):   399.9 ms … 409.0 ms    10 runs

 string-repeat.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/string-repeat.fish
-  Time (mean ± σ):     299.0 ms ±  15.0 ms    [User: 254.7 ms, System: 41.1 ms]
-  Range (min … max):   283.4 ms … 328.9 ms    10 runs
+  Time (mean ± σ):     293.7 ms ±   2.7 ms    [User: 253.0 ms, System: 41.3 ms]
+  Range (min … max):   289.3 ms … 297.0 ms    10 runs

 string-wildcard.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/string-wildcard.fish
-  Time (mean ± σ):     703.1 ms ±  16.1 ms    [User: 462.3 ms, System: 241.3 ms]
-  Range (min … max):   677.8 ms … 728.6 ms    10 runs
+  Time (mean ± σ):     697.0 ms ±   7.2 ms    [User: 454.7 ms, System: 245.8 ms]
+  Range (min … max):   685.2 ms … 706.3 ms    10 runs

 string.fish
 Benchmark #1: release/fish --no-config benchmarks/benchmarks/string.fish
-  Time (mean ± σ):     912.1 ms ±  10.5 ms    [User: 701.3 ms, System: 213.0 ms]
-  Range (min … max):   894.9 ms … 924.5 ms    10 runs
+  Time (mean ± σ):     923.6 ms ±  11.5 ms    [User: 706.2 ms, System: 217.7 ms]
+  Range (min … max):   899.1 ms … 941.3 ms    10 runs

Ignoring the tests that don't show any statistical meaningful results due to not relying on path_get_path_core() that much:

• external_cmds improves by 10%
• load_completions improves by an incredible 77%

[ridiculousfish]

LGTM, nice targeted fix.

[mqudsi]

Thanks. I could have pulled this higher into the call stack where we have a bit more context available regarding the nature of the PATH variable itself but decided to implement it in path_get_path_core() at least for now, where it can be a much simpler change. If there's some use case I'm completely missing here (which is possible even though WSL is my primary driver), we could consider that.