Experiment with changes to the BiMap library
#173Comparing
andir:rust-graph
(fdbe510
) withmaster
(957ff81
)-51%
- Improvements
- 2
- Regressions
- 1
- Untouched
- 1
- New
- 0
- Dropped
- 0
- Ignored
- 0
Benchmarks
Failed
test_deep_layers_large_graphRegression
tests/benchmarking/test_benchmarking.py::test_deep_layers_large_graph-51%
1.3 s
2.7 s
Improved
test_top_level_large_graph
tests/benchmarking/test_benchmarking.py::test_top_level_large_graph+26%
260.8 ms
207.4 ms
test_build_django_from_cache
tests/benchmarking/test_benchmarking.py::test_build_django_from_cache+16%
224.4 ms
194.3 ms
Passed
test_build_django_uncached
tests/benchmarking/test_benchmarking.py::test_build_django_uncached0%
6.2 s
6.2 s
Commits
Click on a commit to change the comparison rangeBase
master
957ff81
-51%
Include the benchmarking string within the binary
This reduces the amount of benchmarking noise that might stem from
noise in filesystem access time.
8a89d00
20 days ago
by andir0%
Remove a potential clone of the get_modules result
Most of the time the result should be cached. Unfortunately we are
required to keep it in a Sync structure (thus a Mutex + Arc is being
used). To workaround this the MutexGuard that we acquire during the
function call is moved into another struct that supports a std::ops::deref-alike
function which returns either a reference to an owned copy (slow case)
or a reference to the value within the Mutex (fast path).
Thus clones or allocations are only done whenever the get_modules
value was invalidated *or* when there is a lock contention (which
shouldn't currently happen due do the GIL).
fdbe510
19 days ago
by andir