Improve directory scanning performance #4426
Conversation
|
Incremental compile with no changes (seconds): |
|
retest uvisor |
|
@mazimkhan Please help with uvisor retest here? |
|
retest uvisor |
|
@0xc0170 There is build failure hence re-running CI will not make it work. Please always have a glance at the logs before re-running CI. |
I did and I considered the failure to be non related to this patch. Looks like I was wrong, this is the second time, it is related. |
|
@mazimkhan |
theotherjimmy
force-pushed
the
faster-scan
branch
4 times, most recently
from
ee53fa0 to
8d4779a
Compare
|
I think this was not ignoring case correctly. Unfortunately, we were depending on "fnmatch.fnmatch" but the code was effectively doing "fnmatch.fnmatchcase". I changed it to do the correct thing. |
|
Huh, different error. |
theotherjimmy
force-pushed
the
faster-scan
branch
2 times, most recently
from
7a07cdf to
f164eb5
Compare
theotherjimmy
force-pushed
the
faster-scan
branch
from
f164eb5 to
85748db
Compare
|
I think I got most of the fat out of the directory scanning. We're down to between 15ms and 45ms spent in resource scanning |
|
|
|
/morph test |
Result: FAILUREYour command has finished executing! Here's what you wrote!
OutputTest failed! |
|
/morph test |
Result: SUCCESSYour command has finished executing! Here's what you wrote!
OutputAll builds and test passed! |
Description
Graph of the first commit compared to master in the next comment.
The remaining commits allow us to correctly elide much processing. In particular, if we delay scanning of features until we try access their respective resource objects, we can avoid scanning some resources by never trying to access their resource objects. Turns out that this optimization saves us ~1/2 of the scan resources time!
Todos