Optimize COPYTREE_* macros.
The main problem slowing down those macros was that it was ending up
calling chmod(1) once for each file and directory.
It was doing this because it was running find(1) in the source tree and
needed to chmod in the target tree, so it had to append the full path to
the chmod(1) calls, which made it impossible to use -exec +.
Rewrite the -exec calls to call sh and cd into it before running
chmod(1) and do so on as many files as we can using the -exec + construct.
While there, optimize a bit more after figuring out that it is possible
to only use one find(1) to run two -exec for different options.
Also switch to using the more powerfull modern regular expressions
instead of the usual basic regular expressions by using find -E. This
will allow the use of | which as no basic regular expressions equivalent
(as in, for example, "*.(html|txt)") as well as the + and ? keywords,
and back references.
About the speed gain, I did a few tests, using time's builtin from csh (table
with all the data available in the review), comparing the home baked versions
of COPYTREE_SHARE found in ports, it boils down to:
version | www/sit (~2k) | www/mathjax (~30k) |
+------------+------------------+--------------------+
home baked | 0.032u 0.104s | 0.946u 7.446s |
current | 0.281u 0.649s | 5.446u 18.626s |
new | 0.024u 0.081s | 0.770u 7.012s |
PR: 213614
Submitted by: mat
Exp-run by: antoine
Sponsored by: Absolight
Differential Revision: https://reviews.freebsd.org/D8283