Check the pattern before anything else, since it doesn't require metadata

[tavianator]

This should partially address #432 by decreasing the number of stat() calls:

$ strace -c -f ./fd-before '\.h$' /usr -j1 -S +1k >/dev/null
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
 15.71    8.831948           7   1192279     46059 stat
$ strace -c -f ./fd-after '\.h$' /usr -j1 -S +1k >/dev/null
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
  7.92    1.972474          10    183907     46046 stat

Though it's not as few as possible:

$ strace -c -f find /usr -iname '*.h' -size +1k >/dev/null
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
 19.01    0.946500           5    161649           newfstatat
$ strace -c -f bfs /usr -iname '*.h' -size +1k >/dev/null
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
 13.73    0.406565           5     69005           statx

Performance is much better when metadata is required:

$ hyperfine ./fd-{before,after}" '\.h$' /usr -j1 -S +1k"
Benchmark #1: ./fd-before '\.h$' /usr -j1 -S +1k
  Time (mean ± σ):      4.623 s ±  0.154 s    [User: 1.465 s, System: 3.354 s]
  Range (min … max):    4.327 s …  4.815 s    10 runs

Benchmark #2: ./fd-after '\.h$' /usr -j1 -S +1k
  Time (mean ± σ):      2.650 s ±  0.058 s    [User: 1.258 s, System: 1.592 s]
  Range (min … max):    2.568 s …  2.723 s    10 runs

Summary
  './fd-after '\.h$' /usr -j1 -S +1k' ran
    1.74 ± 0.07 times faster than './fd-before '\.h$' /usr -j1 -S +1k'

While remaining the same when it's not:

$ hyperfine ./fd-{before,after}" '\.h$' /usr -j1"
Benchmark #1: ./fd-before '\.h$' /usr -j1
  Time (mean ± σ):      2.382 s ±  0.038 s    [User: 1.221 s, System: 1.286 s]
  Range (min … max):    2.325 s …  2.433 s    10 runs

Benchmark #2: ./fd-after '\.h$' /usr -j1
  Time (mean ± σ):      2.362 s ±  0.034 s    [User: 1.193 s, System: 1.294 s]
  Range (min … max):    2.307 s …  2.422 s    10 runs

Summary
  './fd-after '\.h$' /usr -j1' ran
    1.01 ± 0.02 times faster than './fd-before '\.h$' /usr -j1'

[sharkdp]

Thank you very much for your contribution and for the detailed benchmarks. This looks great.

I want to take a more detailed look before I merge this, but this will probably take me a few days.

A small thing I noticed: You are comparing find|bfs /usr -name '*.h' -size +1k with the following fd call:

fd '.h$' -j1 /usr -S +1k

It will probably not make a huge difference, but the . will actually match any character instead of a literal . character. So this will also find .sh files, for example. In addition, fd uses smart-case by default, so it will also match against a capital H as last character. To make the comparison fair, I think you'd have to use the --case-sensitive/-s option and escape the .:

fd --case-sensitive '\.h$' -j1 /usr -S +1k

Even easier, if your are just searching for a file extension, we can use the --extension/-e option:

fd . -e h -j1 /usr -S +1k

[tavianator]

All good points. I updated the commit message with some improved benchmarks (on a different computer), and also moved the extension check up with the pattern check.

[sharkdp]

Thank you very much!