0.8.1 - 2025-01-09
- Miscellaneous code changes and lints
0.8.0 - 2024-12-07
- Bump MSRV to
1.83
- Bump
dactyl
to0.9
0.7.0 - 2024-11-28
- Bump MSRV to
1.81
- Bump
dactyl
to0.8
- Miscellaneous code changes and lints
0.6.1 - 2024-09-27
- Bump
unicode-width
to0.2
0.6.0 - 2024-09-05
- Bump MSRV to
1.77
- Minor code changes and lints
0.5.3 - 2024-02-08
- Bump
dactyl
to0.7
- Miscellaneous doc/script cleanup
0.5.2 - 2023-10-15
- Bump
dactyl
to0.6
0.5.1 - 2023-10-05
- Minor code lints and cleanup
0.5.0 - 2023-06-01
- Bump MSRV
1.70
- CI: test in debug and release modes
- CI: verify MSRV
0.4.0 - 2023-01-26
- Bump MSRV
1.66
- Fix ci badge syntax (docs)
- Remove
black_box
fallback and related build dependency - Remove
BRUNCH_DIR
environmental variable
0.3.7 - 2022-12-15
- Prefer newly stabilized
std::hint::black_box
black_box
fallback (for Rust <=1.65
)
0.3.6 - 2022-09-22
- Improved docs
- Miscellaneous code cleanup
0.3.5 - 2022-09-12
BRUNCH_HISTORY=/path/to/file
env
- Bump MSRV
1.63.0
- Drop
serde
andserde_json
dependencies Bench::with_samples
now requiresu32
rather thanusize
- Bench name lengths are now explicitly limited to
65,535
BRUNCH_DIR=/path/to/dir
env (useBRUNCH_HISTORY
instead)
0.3.4 - 2022-09-09
benches!(inline: …)
macro variant for use inside a custommain()
- Update dependencies
0.3.3 - 2022-09-06
- Improved float handling;
- Minor code cleanup, refactoring;
- Drop
quantogram
dependency (it is now dev-only);
0.3.2 - 2022-08-23
- One final coloration tweak. (Sorry for the release spam!)
0.3.1 - 2022-08-23
- Coloration tweaks;
- Improved float handling;
- Suppress Change summary column when there aren't any;
- Warn if the same
Bench
name is submitted twice;
- env
NO_BRUNCH_HISTORY
should only apply when=1
; - Normalize whitespace in
Bench
names to prevent display weirdness;
0.3.0 - 2022-08-22
This release includes a number of improvements to the Brunch
API, but also some breaking changes. Existing benchmarks will require a few (minor) adjustments when migrating from 0.2.x
to 0.3.x
.
First and foremost, Bench::new
has been streamlined, and now takes the name as a single argument (rather than two). When migrating, just glue the two values back together, e.g. "foo::bar", "baz(20)"
to "foo::bar::baz(20)"
.
Each bench now runs until it has reached either its sample or timeout limit, rather than running as many times as it can within a fixed time period. Existing benchmarks with Bench::timed
will need to switch to Bench::with_timeout
if that was used to extend the run, or removed if used to shorten it.
The execution methods have been cleaned up as well, and now come in three flavors:
Method | Argument(s) | Description |
---|---|---|
Bench::run |
FnMut()->O |
For use with self-contained (argument-free) benchmarks. |
Bench::run_seeded |
I: Clone , FnMut(I)->O |
For use with benchmarks that accept a single, cloneable argument. |
Bench::run_seeded_with |
FnMut() -> I , FmMut(I)->O |
Also for benchmarks that accept one argument, but one that's easier to produce from a callback. |
Bench::run
corresponds to 0.2.x
's Bench::with
, while Bench::run_seeded*
is akin to 0.2.x
's Bench::with_setup*
.
There is no longer any explicit argument-as-reference version, but you can accomplish the same thing using Bench::run_seeded
, like:
Bench::new("hello::world(&15)")
.run_seeded(15, |s| hello::world(&s))
Time-tracking is now done per-method-call (rather than in batches), meaning the precision is now capped at the level of nanoseconds. This improves the results in a number of ways, but means really fast methods won't chart in a meaningful way anymore.
If you need to compare really fast things, it is recommended you perform some sort of iteration within the callback being benchmarked (to increase its runtime). Take a look at the main documentation for an example.
That's it! Apologies for the switch-up, but hopefully you'll agree the new layout is friendlier and more flexible than the old one. :)
- Benches can now be constructed without the
benches!
macro if desired, using the newBenches
struct. Refer to the main documentation for an example; Bench::with_samples
(to set target sample limit);Bench::with_timeout
(to set duration timeout);quantogram
andunicode-width
have been added as dependencies;- The environmental variable
BRUNCH_DIR
can be used to specify a location other thanstd::env::temp_dir
for the history file. - The environmental variable
NO_BRUNCH_HISTORY
can be used to disable run-to-run history altogether.
Bench::new
now accepts name as a single argument instead of two;- Improved statistical analysis, particularly in regards to outlier detection/removal;
- Improved visual display, particularly in regards to multibyte column layouts;
- Improved memory usage for seeded benchmarks;
- Time tracking is now capped at
nanoseconds
;
Bench::timed
(seeBench::with_timeout
);Bench::with
(seeBench::run
);Bench::with_setup
(seeBench::run_seeded
/Bench::run_seeded_with
);Bench::with_setup_ref
(seeBench::run_seeded
/Bench::run_seeded_with
);serde_derive
dependency;
0.2.6 - 2022-06-18
- Update dependencies;
0.2.5 - 2022-04-12
- Enable
num-traits
crate featurei128
(needed for some targets).
0.2.4 - 2022-03-20
Bench::spacer
(for visual separation of results);
0.2.3 - 2022-03-15
- Update dependencies;
0.2.2 - 2022-01-13
- Updated docs;
- Cleaned up lint overrides;
0.2.1 - 2021-12-02
- A demo benchmark (
fn_fib
).
- The list passed to the
brunch::benches!
macro may now include a trailing comma.
0.2.0 - 2021-10-21
- This changelog! Haha.
- Use Rust edition 2021.