This is a benchmark for some different methods of searching for a file/dir
upwards in a filesystem hierarchy. Disclaimer: the code is pretty ad-hoc and
hairy.
Finding a specific file/dir in parent directories is useful as a fast
preliminary check before running more complex and potentially slower commands.
Such use case is checking if directory is tracked in a version control system,
and in this particular case — mercurial. So here all `hgroot_*.sh` files search
for a `.hg/` directory upwards.
Here's the output of `test.sh`:
    not tracked
    builtins 	 dirname 	 upsearch
    ok 0.292	 ok 0.577	 ok 0.207	 
    ok 0.295	 ok 0.700	 ok 0.218	 
    ok 0.303	 ok 0.836	 ok 0.234	 
    ok 0.306	 ok 0.967	 ok 0.261	 
    ok 0.324	 ok 1.147	 ok 0.254	 
    ok 0.271	 ok 1.248	 ok 0.260	 
    ok 0.344	 ok 1.387	 ok 0.261	 
    ok 0.339	 ok 1.555	 ok 0.267	 
    ok 0.327	 ok 1.722	 ok 0.274	 
    ok 0.343	 ok 1.734	 ok 0.277	 
    tracked in mercurial
    builtins 	 dirname 	 upsearch
    ok 0.294	 ok 0.330	 ok 0.203	 
    ok 0.292	 ok 0.466	 ok 0.228	 
    ok 0.305	 ok 0.666	 ok 0.230	 
    ok 0.306	 ok 0.729	 ok 0.225	 
    ok 0.309	 ok 0.848	 ok 0.231	 
    ok 0.316	 ok 0.997	 ok 0.250	 
    ok 0.315	 ok 1.139	 ok 0.242	 
    ok 0.324	 ok 1.202	 ok 0.244	 
    ok 0.314	 ok 1.323	 ok 0.253	 
    ok 0.324	 ok 1.471	 ok 0.274	 
- `builtins` = `hgroot_1.sh`, `dirname` = `hgroot_2.sh` and `upsearch` =
  `hgroot_3.sh`
- results are in seconds x 100, that is, 0.292 is how many a hundred calls to
  the command takes, so each call is ~2.92ms
- `ok` means the command ran correctly
First the test script runs (100 times for a good measure) all contestant
implementations in directories without a `.hg/` anywhere upwards (assuming your
`/` or `/tmp` is untracked), one directory deeper at a time: so the last result
is for the case when the script is 10 directories deep inside `/tmp/tmp.XXX/`
directory (test root). Then the test script adds an empty `.hg/` directory in
the test root and runs everything one more time. It also checks that the
contestants produce the correct answer (path to the directory containing
`.hg/`) and marks the time with either `ok` or `fail`. There are quirks, as I
said, but I believe the results are fairly accurate still.