# Changelog

## 3.2.1
* Trims the README preamble. The old one led with thanks to a package that
  has since been unpublished, and described this one as a slightly modified
  version of it, which stopped being true several releases ago.
* Drops two dead links. `ephemeris-moshier` was unpublished from npm in May
  2026 and `github.com/xErik/ephemeris-moshier` now 404s.
* Keeps attribution, in one line rather than three paragraphs. The licence
  file is unmodified GPL-3.0 boilerplate with the author placeholders never
  filled in, and the ported sources carry no headers, so the README is the
  only place anyone is credited for work this is derived from.
* Documentation only.

## 3.2.0
* Makes the calculation engine pluggable. The built in Moshier engine stays
  the default and nothing about a plain install changes: still no
  dependencies, still pure JavaScript, still the same numbers.
* Adds adapters for two other engines, neither of them a dependency. Install
  `sweph` for Swiss Ephemeris, or `astronomy-engine` for a pure JavaScript
  MIT licensed alternative, then `ephemeris.use(...)`. Both are registered
  whichever is installed, so `ephemeris.backends()` reports what is usable
  here and why not when it is not.
  Note that Swiss Ephemeris is AGPL-3.0 or commercially licensed, which is
  stricter than this package and reaches hosted services. Installing it is a
  licensing decision, which is why it is not installed for you.
* Every backend returns the whole catalogue, so the result shape never
  depends on the engine. A body a backend cannot produce comes back with the
  same keys, every value `null`, `available: false` and a reason that is
  often actionable, rather than throwing or being silently absent.
* Adds `getBody`, for anything the catalogue does not name: numbered minor
  planets, fixed stars, or an engine's own identifiers. Takes one spec or a
  list, and unresolvable entries get the same template.
* Adds `available`, `unavailableReason` and `backend` to every observed
  body, and `backend` to the result. Additive; no existing field changed
  name, type or value.
* Adds `use`, `backend`, `backends` and `registerBackend`, so you can plug in
  your own engine. `src/backends/registry.js` documents the contract.
* Fixes `sirius.apparentLongitudeDd`, which reported right ascension in
  radians while every other body reported degrees. It is now degrees, so
  101.4549 where it used to say 1.7707, and it finally agrees with the
  `101 degrees 27' 17"` string printed beside it. The strings themselves are
  unchanged. Note that for a star the underlying quantity is still right
  ascension rather than ecliptic longitude, which is an older problem and is
  recorded as a known gap.
* Documents how to obtain the Swiss Ephemeris data files, which the previous
  note assumed you already had.
* Adds `npx ephemeris-fetch-ephe` and `ephemeris.fetchEphemeris()` to fetch
  them, about 2.1 MB. Downloads are pinned to an upstream commit and checked
  against a recorded SHA-256, so the same bytes arrive every time and a
  substituted or truncated file is refused rather than written. Running it
  again leaves good files alone.
  `fetchEphemeris({prompt: true})` asks first, naming what it will fetch,
  from where, and under what licence. It only asks when something is
  missing, and off a terminal it refuses rather than hanging or assuming
  yes. Pass `confirm` to answer from your own interface.
  Nothing downloads on install or during a calculation. `getAllPlanets` is
  synchronous, so asking mid calculation would mean blocking the event loop
  or making every signature async, and there is no terminal to ask on in CI
  or in a server anyway. A Swiss Ephemeris body that needs a missing file
  names the command in its `unavailableReason`.

## 3.1.1
* Restores the `3.0.1` changelog entry, which was lost when the lunar node
  work renamed that heading to `3.1.0` on the assumption the two would ship
  together. They did not: 3.0.1 was already released. The entry now sits
  with the version that actually carried it.
* Adds a `2.1.0` entry, reconstructed from the published tarballs. That
  version has been on npm since July 2025 with nothing written about it.
* Adds a test comparing the git tags, which the release workflow creates for
  every publish, against the changelog headings, so a released version
  cannot go undocumented again.
* Documentation and tests only.

## 3.1.0
* Adds the lunar nodes as bodies: `rahu` and `ketu` for the mean nodes, and
  `trueRahu` and `trueKetu` for the osculating ones. They come back from
  `getAllPlanets` like anything else, and `getPlanet` accepts them.
  The mean node was already being computed inside the library and thrown
  away: `gplan` derives the moon's mean longitude and its mean distance from
  the node, and the node is the difference. It agrees with Meeus 47.7 to
  within 0.34 arcsec from 1700 to 2150. The true node comes from the
  orbital plane the moon is on at the instant, and lands within 0.37 arcsec
  of the moon's own longitude at a northward ecliptic crossing, which is the
  definition of the node.
  These are tropical longitudes of date, like everything else here. There is
  no ayanamsa, so `rahu` is not the sidereal Rahu until you subtract one.
  Nothing that already existed changed value.

## 3.0.1
* Moves the development and release instructions out of the README and into
  `CONTRIBUTING.md`, which is not in `files` and so does not ship. npm
  renders the README as the package page, so those sections were showing up
  there as a maintainer runbook in front of anyone looking at the library.
* Moves the changelog into this file. The README keeps a pointer and the one
  note that matters before upgrading. npm has no changelog of its own, it
  only renders the README, so the release notes on GitHub are now the
  per-version view and this file is the full history.
* Documentation only.

## 3.0.0 Breaking Changes - Please read carefully
* Reports `is_retrograde` only for bodies that can actually be retrograde,
  which means the planets and the asteroids. The sun and the moon are never
  overtaken by the earth, so they cannot appear to move backwards; they now
  report `undefined` rather than a permanent `false`, matching what stars
  already did. Their `apparentLongitudeDdPerDay` is unaffected. Sampled over
  four centuries, solar motion never drops below 0.95 degrees a day and
  lunar motion never below 11.76, so there was never an answer to give.
* Removes the unreachable iterative rise and set solver from `transit.js`,
  about 300 lines. It was never called from anywhere, in any commit, and the
  `kepler` call inside it was wrong from the initial 2018 commit, so it threw
  the moment it ran. Nothing changes for callers: the approximate times were
  always the ones being reported. Removing it also drops the last circular
  import in the package, `altaz` and `transit`, which made requiring
  `src/astronomy/moshier/transit` before the package entry point return a
  half built module.
* Fixes rise, set and meridian times falling outside the day they belong
  to. They are measured from the start of the UT day holding the requested
  instant, but the instant itself is terrestrial time, and nothing wrapped
  an event landing either side of that boundary back into the day. So
  `approxSetUT` could report `hours: 43`, or a negative time. Querying at
  exactly midnight always tripped it, which is what `new Date('2026-04-18')`
  gives you. Across a year at six locations, 2179 of 7300 rise and set
  values were 24 hours out and 365 were negative; now none are. The times
  themselves were always right, and still agree with a bisection on the
  library's own elevation to within a few seconds.
* Ships TypeScript definitions in `index.d.ts`. `getPlanet` is generic, so
  its result is narrowed to the body requested.
* Modernises the source to ES2022 declarations. Every `var` became `let` or
  `const`, and `common.copy` takes rest parameters instead of reading
  `arguments`. The calculation itself is untouched and the output is byte
  identical, verified across five dates from 1900 to 2099 including the
  topocentric and transit values.
* Requires Node 22 or newer. The calculation code is untouched and still
  produces byte identical results on older versions, but `engines` now
  states what is actually tested and supported. This matters because Yarn
  refuses to install a package whose `engines` it does not satisfy, where
  npm only warns.
* Node 20 went end of life on 30 April 2026, which is what prompted this.
  Continuous integration now covers Node 22, 24 and 26, and the toolchain in
  `mise.toml` moves to Node 24, the active LTS.

## 2.3.0
* Fixes `is_retrograde`, which was previously wrong for most bodies. It
  compared two reductions of the same position vector rather than two
  positions, so the difference it measured was the change in obliquity over
  an hour, not the body's own motion. The flag was also inverted, missing for
  the sun and the moon, and dependent on the host machine's timezone.
* Adds `apparentLongitudeDdPerDay`, the apparent motion in longitude
* Adds `getBodyNames()`, and makes `getPlanet` throw on an unknown name
* Migrates ESLint to the flat config required by ESLint 9, so `yarn run lint`
  works again
* Replaces the placeholder test with real coverage of positions, retrograde
  windows and timezone independence

## 2.2.0 Feature added
* Adds retrograde flag for planets

## 2.1.0
* Released 26th July 2025
* Drops the `postinstall` build step and the generated `build/` directory,
  so installing no longer runs a script
* Drops the `path` dependency and adds ESLint and a mise config
* Moves `custom.constant.js` under `src/`
* Reconstructed after the fact from the published tarballs. It was tagged
  and published from a commit that is no longer in the repository history.

## 2.0.0 Breaking Changes - Please read carefully
* Released 5th July 2020
* Removes dependency on Moment.js and uses JavaScript's default date object
* Fixes names of a few variables as there were typos (Thanks [Erik](https://github.com/stenflo) for these changes)

## 1.0.0
* Released 30th June 2020
* updates test and fixes test