My personal web garden & blog. http://git.krz.sh/cmc/cleberg.net.git/atom?h=migrate-to-orgo 2026-08-12T01:34:02+00:00 2026-08-12T01:34:02+00:00 Christian Cleberg hello@cleberg.net 2026-08-12T01:34:02+00:00 urn:sha1:ab149c1ae4fe38048fc2d32edac6061e7f3b4638 Remove what the migration left behind, then cut build.py down to the work orgo does not do: 443 lines to 168. - Delete publish.el and theme/templates/, superseded by content/orgo.toml and content/templates/. theme/static/ stays; orgo publishes it as assets. - Drop the tag builder, the org-source copy, and the helpers that only fed them. orgo generates /tags/ and a page per tag from a collection. - Rewrite the README and the migration-era comments that explained the config by pointing at what weblorg used to do. - Serve styles.css as authored. Minifying 1KB of CSS saved 52 bytes over the wire once compressed, which did not pay for a build step and a toolchain dependency. - Stop wiping the output directory. The .orgo-cache.json inside it is what makes a build incremental, and orgo already prunes outputs whose source is gone and re-emits any that are missing. Production keeps .build/ and development gets .build-dev/, because production rewrites image URLs in the output and a shared directory would let one environment reuse pages rendered for the other. An incremental build is byte-identical to a clean one, and the site output is unchanged apart from the /uses/ stack row. 2026-08-12T00:26:42+00:00 Christian Cleberg hello@cleberg.net 2026-08-12T00:26:42+00:00 urn:sha1:5be4598b09a2685d73665b403bf80aad50b15834 .orgo-cache.json lives in the output directory because it describes that directory, and the first orgo deploy put it on the server. Excluding it from rsync stops that, and stops --delete removing it locally between builds. The copy the first deploy left behind is gone from the server. 2026-08-12T00:16:03+00:00 Christian Cleberg hello@cleberg.net 2026-08-12T00:16:03+00:00 urn:sha1:d10ca60103b924b6d182e09123af6ba4a5fffc77 The generator changes; the steps around it that orgo does not do stay. Four of them go, because it does: - the blog index groups itself by year (a template, using page.year) - the tags page is a collection - the recent-posts list on the home page comes from the blog collection - the sitemap is written once site.base_url is set What remains is minifying CSS, copying the org sources for readers who want them, and rewriting img.cleberg.net to /img/ so the onion serves images from its own origin. That last one is why this script still exists. Verified against the live site before deploying: 195 pages, 186 identical in content, and every URL in the live sitemap still produced. The nine that differ are the ones we chose — the reply footer is opt-in now, /tags/ has a page per tag, and footnotes carry aria labels. publish.el and theme/templates/ are left alone. They are weblorg's, and there is no reason to delete a fallback on the same day the replacement goes live. 2026-08-11T21:48:53+00:00 Christian Cleberg hello@cleberg.net 2026-08-11T21:48:53+00:00 urn:sha1:7de79d1a3bf107550edb1defdda72a0008a35e1d The config file follows the tool: content/org-ssg.toml is content/orgo.toml, and the build cache it writes is .orgo-cache.json. Nothing about the site changes — 196 pages, same output. 2026-08-11T21:22:19+00:00 Christian Cleberg hello@cleberg.net 2026-08-11T21:22:19+00:00 urn:sha1:093896152b80da4aac2396eaf32df6c94529f22e The link's closing brackets were followed by a space, so the sentence ended "through time ." on the page. Left over from removing something, most likely. No others like it in the corpus. 2026-08-11T21:18:39+00:00 Christian Cleberg hello@cleberg.net 2026-08-11T21:18:39+00:00 urn:sha1:32cc6cbaa410edec0ab9f8b540f473c1dfd02f40 `[cite/t:@brandur]` needs a citation processor and a bibliography, and there is neither — so weblorg renders it as "?? (????)" on the live page today, and org-ssg leaves the source text visible. The sentence already links Brandur's post immediately after, so the citation was only supplying his name. Now it just says his name. Correct under either generator, and one less piece of machinery the site depends on for one word. 2026-08-11T21:12:19+00:00 Christian Cleberg hello@cleberg.net 2026-08-11T21:12:19+00:00 urn:sha1:3c9514a6070d0fc0c4b2ffda2eef30981f266e54 Routing garden notes and /now/ through post.html put the invitation to reply on them because a directory said so, which is the wrong thing to be automatic — it is a choice about each page, and I had made it site-wide to chase parity with the live site. Those two rules are gone; blog posts keep it. The wording now lives in templates/reply.html rather than inside post.html, so it can be edited or emptied without touching how a post is built, and editing it re-renders the 170 pages that include it and nothing else. Any page opts in for itself with `#+TEMPLATE: post.html`; any post opts out with `#+TEMPLATE: base.html`. 2026-08-11T21:07:17+00:00 Christian Cleberg hello@cleberg.net 2026-08-11T21:07:17+00:00 urn:sha1:a26225589d9cda5328d70d7545e9919d3c0d3f8b - **The feed carries whole posts again.** `include_content = true` on the feed collection, and the template embeds `post.content` rather than the excerpt. 169 items, 2.9 MB, same as weblorg produces. - **Section numbering on.** The live site numbers its headings — inherited from Emacs' default rather than chosen, but switching generators is the wrong moment to quietly redesign 174 pages. The contents list numbers to match. One config line to flip if you decide against it. - **The reply footer follows the live site**: blog posts, garden notes and /now/, but not /uses/, /tips/ or /salary/. Two more `[[pages]]` rules. - **Tags stay one page per tag**, deliberately unlike the live single page with every post inlined under an anchor. A tag with its own URL is worth linking to, and the index stays short. Each index entry carries the tag's slug as an `id`, so old `/tags/#audit` links still land on the right row. Also fixes the garden emoji, which I had transcribed as 🌳 where the weblorg template has 🪴 — taken from that template this time rather than typed. Content now matches the live site on 192 of 195 pages. The three that differ: footnote section markup, an `org-cite` citation neither generator renders usefully, and the tags page above. 2026-08-11T20:31:39+00:00 Christian Cleberg hello@cleberg.net 2026-08-11T20:31:39+00:00 urn:sha1:4d3f7fe4d19b0bff06f53d11f34746b421c7c042 org-ssg can read asset directories outside its source now, so the four files copied in — gpg.txt, robots.txt, salary.csv and styles.css — go back to having one home. `[build] assets = ["../theme/static"]` publishes that directory at /, which is where weblorg serves it from too. That also removes the note asking whoever comes next to keep two stylesheets identical, and picks up styles.min.css as a side effect. 2026-08-11T19:28:56+00:00 Christian Cleberg hello@cleberg.net 2026-08-11T19:28:56+00:00 urn:sha1:584da4a4e0fec8873b3168ec5b0002a1bbc1c392 Two of the differences the page-by-page diff against the live site turned up. The TOC was the larger one: live has one on 171 pages and the org-ssg build had none. `[html] toc = true` only *exposes* `page.toc` — rendering it is the layout's job, and base.html never did. It now emits the same `div#table-of-contents` Emacs does, from a recursive macro over the tree. The anchors already matched, so every link lands where the live one does. Pages that say `#+OPTIONS: toc:nil` get an empty tree and render nothing, which is why /uses/, /now/ and /salary/ have no TOC here either — the same three as live. /blog/ now breaks its list under year headings, as the live index does, using `page.year` and minijinja's groupby. `group_by_year` at the top of blog.html turns it off for one flat list. After both, /blog/ matches the live page's content exactly, and 171 of 171 TOCs are accounted for.