Optimize determine_rpath_dirs #14207
Conversation
There was a problem hiding this comment.
mostly replacing expensive regex with string manipulation.
Can we get effectively the same benefit by checking the prefix and returning early, without rewriting the regex? The code is longer, harder to read and less self-documenting now.
Alternatively what about an lru_cache? This code seems to be run constantly but only have two valid answers: one answer for each for_machine.
Yeah, there's a lot of cleanup possible here:
|
|
Thank you for your good suggestions. I will work on it! |
bruchar1
force-pushed
the
optimize-determine-rpath-dirs
branch
from
1bcd3d1 to
788a761
Compare
|
My profile now tells me 3.4 s instead of 9.2 s for the |
When generating targets in ninja backend for my project, `determine_rpath_dirs` is the second most expensive function call. Here are some optimizations: - Use a single regex instead of 3 to find rpath args - Add cache to get_rpath_dirs_from_link_args - Remove unneeded isinstance check for get_external_link_args - early exit when iterating for rpath arguments - compute system_dirs and external_rpaths once per target instead of recomputing them for every link args of the target (!!!)
bruchar1
force-pushed
the
optimize-determine-rpath-dirs
branch
from
788a761 to
90c8ff2
Compare
When generating targets in ninja backend for my project,
determine_rpath_dirsis the second most expensive function call. Here are some optimizations, mostly replacing expensive regex with string manipulation.