colitur.git/Makefile, branch main

fix(install): ship the OF runtime data files, guard the installed layout

2026-08-26T10:55:23Z

data/dune's install stanza named only the four EF runtime files, so a `dune install`ed colitur exited 2 on any --rite of invocation (`failed to load .../data/of/calendar-2002.sexp`) -- the branch's headline feature was dead on an installed binary. Add all 15 OF files (calendar-2002.sexp, the 13 amendments/*.sexp, lectionary.sexp) to the (files ...) stanza, following the existing ef/ entries' form and the stanza's own stated rule (deliberately not the litcal allow-list, a test fixture, same reasoning already given for the EF pair). Add a regression guard for the class of bug, not just this instance: `make check-install` performs a real `dune install --prefix ` and runs both rites against the INSTALLED binary, asserting exit 0 and the expected day count. test/cli.t's cram sandbox binds %{bin:colitur} inside the build tree, where data_dir()/of_data_dir()'s build-tree fallback candidate resolves even when the installed one is missing entirely -- so no cram test could have caught this. Wired into `make check` (unlike check-schema/check-templates/check-citations, which skip when an optional external tool is absent) since installing needs nothing beyond what building already needs. Verified the guard has teeth: reverted data/dune to the pre-fix version and confirmed check-install fails loudly naming the missing file.

build: verify the upload instead of assuming it

2026-08-21T08:36:00Z

release-tarball chained its steps with ';', so a failed scp or ssh still reached the final echo and it announced a publication that had not happened -- the file was never in the docroot. Every step is now chained with && and the result is checked over HTTP before anything is claimed: the URL must respond and the bytes must untar. Also passes -o ClearAllForwardings=yes, since the i2p port forwards in the ssh config fail when another session holds them and their noise buries real errors.

build: publish the latest tag, not strictly HEAD

2026-08-21T08:35:06Z

Requiring HEAD to be exactly a tag made the target unusable: the commit that adds the download link to the README necessarily comes after the tag it links to. It now archives the most recent tag and says so when HEAD is ahead of it.

build: publish one current tarball instead of per-tag snapshots

2026-08-21T08:34:50Z

cgit offers snapshots for every tag or for none, so colitur.git now has cgit.snapshots=0 and 'make release-tarball' uploads the tagged archive to a stable URL. Ten tags had meant twenty download links, nineteen of them stale. The target refuses to run from an untagged or dirty tree, so what is published always matches a tag.

feat(templates): retune ordo.tex to ordo.typ's density, fix ordo.ms's own

2026-08-20T10:47:47Z

ordo.tex was 53 pages / 6.9 days-per-page on 2027 against ordo.typ's 27 pages / 13.5 -- the user prefers the tighter Typst booklet. Read ordo.typ's own typographic decisions (base size, the day-number/name/detail-line size hierarchy, box padding, inter-box skip, leading) and matched them: an explicit 9pt/10.5 base (was \normalsize, 10pt), the day number and feast name both 9pt bold (bold distinguishes them, not size, as ordo.typ decided), 7pt weekday, 6.5pt detail lines, sharp-cornered boxes, and needspace amounts re-measured for the shorter boxes. Result: 29 pages / 12.6 days-per-page -- within the "a page or two of 27" target, reached without sacrificing legibility (verified against the rendered PDF, not assumed). The one real gotcha: \fontsize set inside a brace group that closes before the line-ending \\ has no effect on the interline glue TeX inserts for that break -- the glue uses the OUTER scope's \baselineskip, not the just-closed group's. This is exactly the shape of bug the pre-retune box already had (every \fontsize group closed before its own \par), which is why the first retune pass barely moved the page count. Fixed by declaring \fontsize/ \selectfont/\bfseries/\mdseries inline, ungrouped, immediately before the text each governs. Recorded in the file's own header for next time. ordo.ms looked comparatively fine (43 pages) but was rendering on A4, not A5 -- roughly double an A5 page's area, so normalised for area it was already looser than ordo.tex's own pre-retune baseline. Fixed, each found by rendering and reading the real PDF rather than assumed: added -P-pa5 to the Makefile's groff invocation (mirroring grid.ms's existing -P-pa4l); fixed the day box reading the RAW rank/colour strings (class-2, green) instead of the resolved Latin pair (rank_name/colour_name) -- exactly the "printed slugs" defect the ordo rebuild's own commit message named, which this file had regressed to; added -Kutf8 (groff's default input encoding produced visible mojibake on an accented Latin name); tightened fonts/ leading the same way as ordo.tex; found and fixed a stray-blank-line bug present since the file's original authoring (every day/week/month boundary left an accidental blank template line, read by groff as its own extra .sp); found and fixed a genuine print-safety bug (text sitting 0.13mm from the true page edge -- .pl, groff's text page length, is independent of -P-pa5's device paper selection and needed matching explicitly before ms's own FM-based bottom margin would fire at all); and added a .ne guard so a week header can no longer be stranded alone at a page's foot. Net: 43 pages (A4, broken) to 39 pages (real A5, print-safe, real Latin names, correct encoding, zero orphans) -- no numeric target set for this flavour, but no lower number was reachable without leaving one of the above unfixed. ordo.html and every grid.* flavour are untouched, as scoped. Both golden/ordo-2027.{tex,ms} regenerated through the real test harness path (Test_support's loaders + Test_view's en-with-la-fallback view_of), not the CLI, whose own --lang default (la) differs from the harness's and would have pinned content the suite never produces. Full suite green: 503 tests (504 with the exhaustive sweep). make check-templates/check-schema/check-citations all clean.

feat(templates): ordo booklet and wall calendar in Typst

2026-08-20T08:07:37Z

Adds templates/ef/ordo.typ (A5, one week per page, each day framed with a colour swatch, real names, every fixed string through term.*) and templates/ef/grid.typ (A4 landscape, one month per page, 7 columns, colour-filled cells, Epistle/Gospel sigla), matching what ordo.tex and grid.tex already do. Typst resolves its own table of contents inside a single compile: no pageref/aux-file dance, unlike pdflatex's own two-pass need for ordo.tex. ordo.typ's TOC is one flat heading per week (Month . Week N) rather than ordo.tex's own month-once/weeks-nested layout, since the view model has no first-of-month flag for a template to test and inventing one would be new engine state for a cosmetic grouping. grid.typ deliberately diverges from grid.tex's own partial-fill behaviour for a 5-week month: Typst's row-size list repeats its own last entry for every remaining row, so a two-entry rows spec fills the page completely regardless of week count, where LaTeX's fixed cellh leaves a 5-week month's bottom sixth blank on purpose. Both divergences are documented in the templates' own headers. Two bugs found and fixed while building these against the real typst 0.14.2 binary: two stray hash characters in grid.typ's header comment that were not actually inside a line-comment (would have been parsed as Typst code, not prose); and two lines merging into one wherever they were separated only by a bare newline (still the same soft-wrapped paragraph in Typst, unlike LaTeX's own explicit paragraph break) -- fixed with a trailing backslash to force a real line break without adding inter-paragraph spacing, verified against a full-year render (zero typst warnings across all 365 days of 2027, in both templates). Goldens (test/golden/ordo-2027.typ, test/golden/grid-2027.typ) were generated through the harness's own View.of_days/Template.render_string path, via a throwaway generator, NOT through the CLI: the CLI's default language table differs from the harness's own English-chained-to-Latin table, confirmed live (diffing CLI output against the harness's own render showed exactly the expected language divergence and nothing else) -- the same trap this project has hit before. make check-templates now typesets both templates with the typst binary, using the same SKIPPED-and-exit-0 shape the pdflatex/groff blocks already use when their own tool is absent, verified in both directions: absent (PATH without typst) skips loudly and exits 0, and a deliberately corrupted template fails the target with a non-zero exit and typst's own error text.

docs(naming): the config man page, and the field reference

2026-08-20T07:28:20Z

colitur-config.5 documents the file's location, the flag > config > default precedence, every setting, and that an unknown key is warned about rather than fatal. Its worked example is a real run against the shipped example overlay, pasted rather than hand-typed -- including one genuine gotcha found while verifying it: overlay/template values are not shell-expanded, so a bare "~/..." written into the config file fails to load; the full path must be written out instead. colitur-templates.5's field reference is updated for the view change -- every field checked against lib/render/view.ml and schema/day-v1.json, not transcribed from memory or an earlier version of this page -- and its worked example is pasted from a real `colitur table` run. Its scope-shadowing warning about {{name.la}} is REMOVED and replaced with a short note that the hazard is now unrepresentable, because name is a plain resolved string with no dotted path left to fall back through. The num collision between a month and a week (both carry num, and a bare {{num}} read from inside a day silently climbs to the week's own value, not the month's) still exists and is demonstrated with a real render, not merely asserted. A man page describing a hazard that was fixed is as wrong as one that omits a real hazard. Two further stale mentions of the removed {{name.la}} hazard, and two places describing the schema as "Latin and English names" rather than one resolved display name, were found in colitur.1 while cross-checking and fixed alongside the primary rewrite; its own CSV example in EMIT was also still the pre-Task-5 column list (name_la/name_en, no *_name columns) and is now the real, verified header. lang/ already installs beside templates/ and schema/ (a prior hotfix); colitur-config.5 is added to install/uninstall/man/doc in the Makefile so `make install`/`make doc` cover it too.

fix(install): ship lang/ so installed colitur resolves names

2026-08-19T21:52:05Z

la.ini and en.ini (725 Latin/English liturgical names -- the entire point of the naming-and-config work just merged) had no dune install rule, so make install / dune install never copied them to /share/colitur/lang/. bin/main.ml's lang_dir() probes exactly that installed path first, falling back to a build-tree layout that also does not exist for an installed binary -- so every installed user silently got raw slugs (ef-circumcision) and untranslated terms (epistle/gospel) instead of resolved names, with no error at all. The defect existed because nobody had run the installed binary from a directory with no colitur source tree above it: dune exec and dune test always resolve lang/ via the build tree, materialized as a side effect of test/dune's cram deps, so every in-tree check stayed green while the installed artifact was broken. Add lang/dune, following schema/dune's exact pattern: an install stanza placing la.ini and en.ini at /share/colitur/lang/, matching where bin/main.ml's lang_dir() already looked. No code change was needed there -- only the missing install rule. Update Makefile's install target help text and summary echo to mention lang alongside ef/templates/schema. Verified end to end: uninstalled, removed lang/dune, rebuilt and reinstalled to reproduce the bug from an outside directory (~/tmp, colitur table showed ef-circumcision/epistle); restored lang/dune, reinstalled, and confirmed the same command now shows the resolved Latin name and rubric terms, that ~/.local/share/colitur/lang/ contains both files, and that colitur emit --format csv shows a real name in the name column. lang/la.ini and lang/en.ini are untouched. Full suite: dune test --force, 495 tests, 0 failures.

docs(templates): document the ordo booklet's required two-pass build

2026-08-19T20:00:09Z

A LaTeX table of contents needs two pdflatex passes -- the first pass leaves every entry showing '??', the second resolves the \pageref values. A user hit exactly this and reported it as a bug; the template was fine, the instructions were not. README's rendering example now runs pdflatex twice for ordo.tex (or names latexmk -pdf as the one-shot alternative) and notes the wall calendar needs only one pass, having no cross-references of its own. make check-templates now runs pdflatex twice per LaTeX template too, so the target exercises what a user actually has to do rather than silently passing on a single, incomplete pass.

fix(tools): harden check_citations.py against its own self-poisoning bug

2026-08-19T13:01:18Z

Reproduced the defect: reintroducing the exact historical citation bug (pointing class-1's citation back at LT.txt:12459, the value a prior fix round corrected away from) made the tool report "147 citations checked, 0 look wrong". The mechanism was that the corrective comment documenting the old bug quotes the wrong historical value, and the checker pooled every quoted phrase from the whole surrounding comment block, so citing the wrong line matched the comment explaining why it was wrong. Four changes: 1. The word pool for a citation is now scoped to the entry(ies) it is attached to only -- never to quoted text elsewhere in the comment. This is the direct fix for the self-poisoning bug. 2. A citation whose pool has fewer than two distinctive words (Latin liturgical headings are short and stopword-heavy) cannot discriminate the right line from a wrong nearby one. Such a citation is now reported CANNOT VERIFY and fails the target, instead of silently passing. 3. The blanket +-2-line tolerance is gone. A bare "LT.txt:N" is checked at line N only; a heading that genuinely wraps must say so explicitly as "LT.txt:N-M". The allowance moves into the data, where it is visible. 4. The tool gets its own test suite, tools/test_check_citations.py, with a synthetic fixture covering: a correct citation, off-by-one and off-by-three mismatches, an explicit wrap range, a degenerate pool, a PATTERN-marked entry with no citation, and a dedicated regression test for the self-poisoning case itself. Wired into `dune test` via a new (rule (alias runtest) ...) in tools/dune (a plain (test ...) stanza cannot run a Python script), so it runs with the rest of the suite, not only as a `make` target. Added a --file/--lt-file override to check_citations.py so the tool (and its own tests) can point at a fixture without touching the real lang/la.ini or docs/research/LT.txt. Confirmed the "SKIPPED, exit 0" behaviour for a missing docs/research/LT.txt is unchanged. tools/__pycache__/ (a stray artefact of this script, previously untracked and ungitignored) is now in .gitignore. Measured against the current lang/la.ini (another task is still landing its sanctoral entries on this branch): 15 of 275 citations now look wrong and 42 more cannot be verified, both far above the 0 the unhardened tool reported. Not fixed here -- the data pass is separate, once the sanctoral entries land.