Merge pull request 'docs(readme): add benchmark table pinned to v0.1.0' (#1) from docs/performance into main

Reviewed-on: https://git.bnema.dev/bnema/kvconf/pulls/1

commit: fff2e749e927298242208bf68a51407ccfd43507
author: bnema · 12d ago · 2026-10-10 06:20 UTC
merge of: 3b1b6ef + 150eab2 · diff against the first parent
signature: SSH UnverifiedSHA256:VOCA5lYARHcbKDZ7GhTV45VmcjWnJ6ZQIBazj17LivA The key is unknown

1 file · +23 -1

M README.md +23 -1


@@ -21,7 +21,8 @@ output.DP-1.scale = 1.5
21	21		- Safe for real configs: atomic saves, symlinked dotfiles, editors that
22	22		save by rename, missing directories, broken files that keep the last good
23	23		config, and warnings that never print values (they may be secrets).
24		-	- Fast: about 800 MB/s to parse, a few allocations per reported change.
	24	+	- Fast: about 750 MB/s to parse, 2 allocations per reported change
	25	+	(see [Performance](#performance)).
25	26
26	27		```sh
27	28		go get github.com/bnema/kvconf

@@ -107,6 +108,27 @@ skips saves that change no value.
107	108		- `WriteFile`: atomic replace, written through symlinks.
108	109		- `ReadFile`: bounded read that rejects FIFOs and devices without blocking.
109	110
	111	+	## Performance
	112	+
	113	+	Measured at `v0.1.0` (`3b1b6ef`, 2026-10-10) with Go 1.27.1 on linux/amd64
	114	+	with an AMD Ryzen 9 7900X3D. Values are medians of 6 runs. The input is a
	115	+	realistic ~2.9 KB config of 100 lines with comments, 60 binds and 8 outputs.
	116	+
	117	+	\| Benchmark \| Time/op \| Throughput \| Bytes/op \| Allocs/op \|
	118	+	\|---\|---:\|---:\|---:\|---:\|
	119	+	\| `Parse` (file to entries) \| 3.9 µs \| 755 MB/s \| 10.4 KiB \| 5 \|
	120	+	\| `Unmarshal` (file to struct, maps included) \| 12.5 µs \| 232 MB/s \| 21.7 KiB \| 47 \|
	121	+	\| `Decode` (5 scalar entries) \| 177 ns \| \| 112 B \| 1 \|
	122	+	\| `Marshal` (struct to file) \| 11.6 µs \| \| 15.5 KiB \| 19 \|
	123	+	\| `Set` (edit one key in place) \| 2.4 µs \| \| 10.5 KiB \| 5 \|
	124	+	\| `Watch` (one write to one report, inotify) \| 5.5 µs \| \| 256 B \| 2 \|
	125	+
	126	+	Reproduce with:
	127	+
	128	+	```sh
	129	+	go test -run '^$' -bench . -benchmem -count 6 .
	130	+	```
	131	+
110	132		## License
111	133
112	134		MIT


@@ -21,7 +21,8 @@ output.DP-1.scale = 1.5
21		- Safe for real configs: atomic saves, symlinked dotfiles, editors that	21		- Safe for real configs: atomic saves, symlinked dotfiles, editors that
22		save by rename, missing directories, broken files that keep the last good	22		save by rename, missing directories, broken files that keep the last good
23		config, and warnings that never print values (they may be secrets).	23		config, and warnings that never print values (they may be secrets).
24	-	- Fast: about 800 MB/s to parse, a few allocations per reported change.	24	+	- Fast: about 750 MB/s to parse, 2 allocations per reported change
			25	+	(see [Performance](#performance)).
25			26
26		```sh	27		```sh
27		go get github.com/bnema/kvconf	28		go get github.com/bnema/kvconf

@@ -107,6 +108,27 @@ skips saves that change no value.
107		- `WriteFile`: atomic replace, written through symlinks.	108		- `WriteFile`: atomic replace, written through symlinks.
108		- `ReadFile`: bounded read that rejects FIFOs and devices without blocking.	109		- `ReadFile`: bounded read that rejects FIFOs and devices without blocking.
109			110
			111	+	## Performance
			112	+
			113	+	Measured at `v0.1.0` (`3b1b6ef`, 2026-10-10) with Go 1.27.1 on linux/amd64
			114	+	with an AMD Ryzen 9 7900X3D. Values are medians of 6 runs. The input is a
			115	+	realistic ~2.9 KB config of 100 lines with comments, 60 binds and 8 outputs.
			116	+
			117	+	\| Benchmark \| Time/op \| Throughput \| Bytes/op \| Allocs/op \|
			118	+	\|---\|---:\|---:\|---:\|---:\|
			119	+	\| `Parse` (file to entries) \| 3.9 µs \| 755 MB/s \| 10.4 KiB \| 5 \|
			120	+	\| `Unmarshal` (file to struct, maps included) \| 12.5 µs \| 232 MB/s \| 21.7 KiB \| 47 \|
			121	+	\| `Decode` (5 scalar entries) \| 177 ns \| \| 112 B \| 1 \|
			122	+	\| `Marshal` (struct to file) \| 11.6 µs \| \| 15.5 KiB \| 19 \|
			123	+	\| `Set` (edit one key in place) \| 2.4 µs \| \| 10.5 KiB \| 5 \|
			124	+	\| `Watch` (one write to one report, inotify) \| 5.5 µs \| \| 256 B \| 2 \|
			125	+
			126	+	Reproduce with:
			127	+
			128	+	```sh
			129	+	go test -run '^$' -bench . -benchmem -count 6 .
			130	+	```
			131	+
110		## License	132		## License
111			133
112		MIT	134		MIT