ls: allocate less strings when formatting item names and hyperlinks

[Nekrolm]

Was looking into #10662

While most time spent in getdents & lstat syscalls, allocations are still taking around 3-5% of the time

[sylvestre]

could you please share the hyperfine benchmark results ? (without the patch, with and with gnu)
thanks

[Nekrolm]

dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ sudo hyperfine --warmup 3 -N -i  "target/release/coreutils ls -R /proc/ > /dev/null" "./coreutils_main ls -R /proc/ > /dev/null" "ls -R /proc/ > /dev/nul
l" 
Benchmark 1: target/release/coreutils ls -R /proc/ > /dev/null
  Time (mean ± σ):     362.8 ms ±  15.2 ms    [User: 138.4 ms, System: 217.8 ms]
  Range (min … max):   348.1 ms … 399.4 ms    10 runs
 
  Warning: Ignoring non-zero exit code.
 
Benchmark 2: ./coreutils_main ls -R /proc/ > /dev/null
  Time (mean ± σ):     382.8 ms ±  24.1 ms    [User: 150.1 ms, System: 226.6 ms]
  Range (min … max):   356.4 ms … 419.2 ms    10 runs
 
  Warning: Ignoring non-zero exit code.
 
Benchmark 3: ls -R /proc/ > /dev/null
  Time (mean ± σ):     297.0 ms ±  26.5 ms    [User: 110.6 ms, System: 184.5 ms]
  Range (min … max):   255.7 ms … 351.4 ms    10 runs
 
  Warning: Ignoring non-zero exit code.
 
Summary
  ls -R /proc/ > /dev/null ran
    1.22 ± 0.12 times faster than target/release/coreutils ls -R /proc/ > /dev/null
    1.29 ± 0.14 times faster than ./coreutils_main ls -R /proc/ > /dev/null

dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ ls --version
ls (GNU coreutils) 9.4
Copyright (C) 2023 Free Software Foundation, Inc.

I don't expect much from these changes.
It has more potential to do better allocation-wise, but with more significant changes (e.g. refactor output width computation to not allocate by formating each visited inode).

But maybe optimising calls to getdents (tweak the buffer size?) is a more promising way.

[Nekrolm]

getdents buffer is not a case, actually:

strace -o gnu.trace  ls -R /proc/ > /dev/null 
strace -o coreutils.trace ./coreutils_main ls -R /proc/ > /dev/null 

dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ wc -l gnu.trace 
90005 gnu.trace
dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ wc -l coreutils.trace 
104756 coreutils.trace

dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep getdent coreutils.trace | head -1
getdents64(3, 0x575ac6349440 /* 352 entries */, 32768) = 9408
dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep getdent gnu.trace | head -1
getdents64(3, 0x57444ada0fd0 /* 352 entries */, 32768) = 9408

gnu version just makes 60% less calls for fstat (newfstatat)

dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep fstat gnu.trace | wc -l
13279
dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep fstat coreutils.trace | wc -l
39813

dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep fstatat coreutils.trace | wc -l
26539
dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep fstatat gnu.trace | wc -l
0

dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep getdents gnu.trace | wc -l
26564
dmis@dmis-asus-N7600PC:~/WORKSPACE/coreutils$ grep getdents coreutils.trace | wc -l
26564

fstatat calls -- around 10% on the flamegraph I posted above

[sylvestre]

please add comments explaining why you are doing this :)

[Nekrolm]

comments are already there for this trait -- updated it

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/tail/inotify-dir-recreate (passes in this run but fails in the 'main' branch)

[sylvestre]

please run rustfmt

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/misc/tee (passes in this run but fails in the 'main' branch)
Skipping an intermittent issue tests/tail/inotify-dir-recreate (passes in this run but fails in the 'main' branch)