Commit `b8df7a53fd`

b8df7a53fd037322fe4b57e2bad5819bd704cc19

parent: 0eff57ddf2

Verified · cmc

cmc <hello@cleberg.net> · 2026-08-11 23:11 UTC

Add the build-script row, and re-measure everything together

The first numbers were taken twenty minutes apart, and it showed: the build
script came out *faster* than the raw emacs run it wraps, which cannot be true.
All four rows are now measured back to back in one session.

The new row is the point of the exercise. weblorg alone does not group a blog
index by year, write a tags page, rewrite image URLs, minify CSS or emit a
sitemap — so ~600 lines of Python do those on top of it, and the whole publish
is paid again on every run. 49.0s becomes 50.3s.

orgo does three of those five natively. Minification is a build step either way,
and orgo has no sitemap at all — recorded in the README as a gap rather than
quietly left out of the comparison.

Deployment is excluded from every row: it is an rsync over someone's network,
which measures the network.

Layout: unified · split

README.md +18 −8

	@@ -82,21 +82,31 @@ hundreds of them stays instant.
82		82
83	## Speed	83	## Speed
84		84
85	Publishing one real site — 178 org files, ~180 pages — with the generator it uses today	85	Publishing one real site — 178 org files, ~180 pages — three ways. Median of three runs
86	and with orgo. Median of three runs each, same machine, same moment:	86	each, measured back to back on one machine:
87		87
88	\| \| Time \| \|	88	\| \| Time \| \|
89	\|---\|---\|---\|	89	\|---\|---\|---\|
90	\| weblorg (`emacs --script publish.el`) \| 50.1s \| \|	90	\| weblorg (`emacs --script publish.el`) \| 49.0s \| \|
91	\| orgo, cold build \| 0.21s \| 239× faster \|	91	\| the same, plus the build script wrapped around it \| 50.3s \| \|
92	\| orgo, nothing changed since last build \| 0.13s \| 385× faster \|	92	\| orgo, cold build \| 0.22s \| 223× faster \|
93		93	\| orgo, nothing changed since last build \| 0.13s \| 377× faster \|
94	Read that with three things in mind. The weblorg figure includes Emacs starting up and	94
		95	That middle row is the interesting one. weblorg alone does not group a blog index by year,
		96	write a tags page, rewrite image URLs, minify CSS or emit a sitemap — so the site's owner
		97	wrote ~600 lines of Python to do those on top of it, and pays for the whole publish again
		98	every time. orgo does the first three natively, minification is a build step either way,
		99	and it has no sitemap yet — that one is a real gap, not a win.
		100
		101	Read the numbers with three things in mind. The weblorg figures include Emacs starting and
95	loading its packages, which you pay on every publish and cannot avoid. orgo emits 13 pages	102	loading its packages, which you pay on every publish and cannot avoid. orgo emits 13 pages
96	weblorg does not (one per tag), so it is doing slightly more work. And the two do not	103	weblorg does not, one per tag, so it is doing slightly more work. And the two do not
97	produce byte-identical output — the differences are deliberate and listed under	104	produce byte-identical output — the differences are deliberate and listed under
98	[Org support](https://krazywarez.github.io/orgo/guide/05-org-support.html).	105	[Org support](https://krazywarez.github.io/orgo/guide/05-org-support.html).
99		106
		107	Deployment is excluded from every row: it is an rsync over someone's network, which
		108	measures the network.
		109
100	Apple M2 Pro, 12 cores, macOS 26.6, Emacs 30.2, orgo built with `--release`.	110	Apple M2 Pro, 12 cores, macOS 26.6, Emacs 30.2, orgo built with `--release`.
101		111
102	## The documentation	112	## The documentation

Commit b8df7a53fd

Commit `b8df7a53fd`