WebAssembly rather than a native Node addon: one artifact serves Node,
Deno, Bun, a browser and anything built on them, where napi would have
meant a prebuilt binary per platform and a source build for whoever the
matrix missed. The engine is pure computation and the menologion is
already compiled in, so there is nothing for it to give up.
A day comes back as a plain object, not a handle. A wasm class would
have meant a .free() that every caller eventually forgets, and this way
JSON.stringify works and the thing spreads into state. Two serializer
defaults had to be overridden to keep the promise that it is the same
document --json emits: maps as objects, and an absent field as null
rather than undefined, which stringify would otherwise drop entirely.
The TypeScript is hand-written, because the generated types for a value
crossing as JsValue are `any`, and a binding that types its whole
return as `any` is a binding TypeScript cannot check. tsc --strict is a release
gate now, the way mypy --strict already is, and the check includes three
lines that must fail to compile.
The one place this differs from every other binding: no remembered
default jurisdiction, because a browser has nowhere to keep one. It is
called out in the README, in the types and in docs/versioning.md.
The test resolves typikon through a staged node_modules rather than a
relative path, so what it exercises is the exports map, and it compares
a year of days field by field against the command line for four
jurisdictions.
An adversarial review of the release, and four of its findings were things
PyPI never lets you replace once uploaded.
- rust-version said 1.74. clap has needed 1.85 since 4.6, and cargo
hard-errors rather than warning, so the sdist advertised a floor it could
not build on. docs/releasing.md sends macOS and Windows users to that
sdist.
- The wheel was classified Development Status :: 4 - Beta on a release the
changelog calls stable, with none of the Python version classifiers PyPI
indexes on.
- Six README links were relative. PyPI resolves them against the project
page, so every one 404'd there, as they do on the live 0.2.0 page now.
- in that order packs the wheel into the
sdist. The tarball on disk was 431KB with an x86-64 binary inside it.
Also: twine upload gains --skip-existing, so a retry after a partial upload
finishes rather than dying on a version that cannot be reused; twine check
gains the renderer that makes it parse the description at all, without which
it returns None and reports PASSED; and the README no longer claims a
twenty-year sweep the test did not run, because 366 * 21 / 4 is five years.
The test now runs the twenty-one it was named for.
The crates.io name belongs to an unrelated project, which the README now says
rather than leaving a reader to find out from cargo.
Nobody depends on this yet, which makes now the moment to commit to a surface
rather than later when someone does.
docs/versioning.md says what that surface is. A calendar has a wider one than
a library usually does: the JSON shape, the fasting level names and their
order, the jurisdiction keys, the command line, and the two language APIs are
all things a caller can reasonably build on, and only one of them is a
function signature. The text output, the reason strings and the menologion's
wording are explicitly not.
It also names the category semver has no word for. A correction to a fast rule
or a feast rank breaks no interface, so it is a patch; but it changes what the
program tells someone to eat. Those are called out in the changelog under
Calendar whatever the version does, and CHANGELOG.md starts by listing every
one of them against pre-1.0 behaviour.
daysToPascha is documented as signed, counting to the Pascha of that day's own
civil year, which it always was and nothing said.
The engine is the same; only the front door is new. pyo3 behind an optional
feature, maturin for the wheel, and the bindings mirror DayJson rather than Day
because DayJson is already the shape front ends read and the one that has to stay
still. to_dict() therefore gives exactly what --json gives, and a test holds the
two to that.
That test earned its place immediately. The bindings were calling
jurisdictions::resolve(None), which answers with a built-in default that knows
nothing about the machine, so Python was reckoning by the New Calendar while the
command line on the same box reckoned by ROCOR — every field downstream of the
calendar differed, including the fast. They now take the same two steps main.rs
takes, and agree field for field on Pascha, the Nativity and Theophany.
The wheel is abi3, so one build serves 3.9 upward rather than one per
interpreter. That costs the datetime C API, which the limited API does not
expose, so a date is read through its attributes instead — which incidentally
makes datetime.datetime work as well as datetime.date. Constructing dates is
unaffected, pyo3 calls Python's own constructor for that.
panic = "abort" comes out of the release profile. An extension module has to be
able to unwind a panic into a Python exception; aborting would take the
interpreter down with it. The CLI keeps the same behaviour it always had, a
panic there just unwinds now.
Mixed layout so the wheel can carry py.typed and a stub file, which a bare
extension module has nowhere to put. mypy --strict is clean against the stubs
and catches a deliberate error, so the Typing :: Typed classifier is not a claim
about nothing.
Python was never the house language here: the mw-* desktop surface it came
from is 39 bash scripts and was exactly one Python script, the one this was
extracted from. What it cost was distribution, a widget dependency that
needed pipx and a Python floor, and 24ms of interpreter startup before a line
of the program ran.
The port is verified rather than asserted. Both implementations were run over
199,446 day-records, 26 jurisdictions across 21 years, with every field of
every record required to match; the JSON, the iCalendar and 480 renderings of
the text output are identical too. examples/dump.rs is the harness.
What the type system actually buys, which is less than the usual claim and
more than nothing:
- Level is an ordered enum, strictest first, so reads as 'a is
stricter' and relaxed_to() is the one place a feast can lift a fast. No
rule can get the direction backwards.
- Date is a newtype over the Julian Day Number and JulianDate is a separate
type, so a civil date and an ecclesiastical one cannot be mixed up. That
confusion is the single most common way calendar code goes wrong.
- Octets is a newtype in the iCalendar folder, because the bug it replaced
counted characters against a limit RFC 5545 states in octets.
- Movables is a struct, not a map, so a mistyped key is a compile error
rather than a None at the far end of the program.
Of the six defects an adversarial review found in the Python, exactly one had
a shape a type system would have caught. The rest were logic, and the tests
came with them.
The menologion is compiled in with include_str!, so the binary answers on its
own: 930KB, no runtime, 1.2ms to start against Python's 36ms.
Feasts, seasons and the fast for twenty-three jurisdictions, derived from
the paschalion and the fixed calendar so it needs no network and cannot go
stale.
The three things calendar code usually conflates are kept apart: the
paschalion that gives Pascha, the menologion the fixed feasts are read on,
and the fasting rule that is kept. Almost every Orthodox Church keeps the
Julian Paschalion; what divides them is the menologion, so one file of
commemorations indexed by ecclesiastical date serves old and new calendars
alike.
Six fasting levels rather than three, because the difference between
xerophagy and wine-and-oil is the difference the books actually make. Level
and obligation are carried separately: the Orthodox rule reports everything
as customary, and Ruthenian particular law reports both how a day is kept
and the far smaller minimum that binds.
The menologion covers all 366 days and is dedicated to the public domain.
Sets published by someone else are fetched, never redistributed.
funkpower