GitHub repositories

The question. Which public repositories contain substantive solar eclipse computation, what do they implement, from which formulas and elements, with which constants, and do any of them handle limb profiles, terrain or catalogue generation?

Inventory scope

The inventory covers thirty public eclipse-computation repositories documented on 2026-09-15. Repository metadata and author documentation identify languages, licences and capabilities. Star counts and last-push dates describe that snapshot. Image-processing software and unrelated products named Eclipse are outside the scope ⁷.

Repositories

Besselian element consumers

RHerAle/Eclipse-Engine. JavaScript, AGPL-3.0, 0 stars, last push 2026-09-14. The engine behind eclipseradar.com for 2026 to 2050: "Besselian shadow geometry: contacts, obscuration, magnitude, central line, limits, shadow outline, and obscuration bands", plus spectral irradiance with limb darkening and ICNIRP exposure limits, "Terrain analysis: skyline toward the Sun from elevation models and OpenStreetMap building data", and solar-limb video stabilisation. Its data/eclipses.json is "the catalogue of Besselian elements, fitted to the JPL DE440s ephemeris", The repository consumes these independent elements. Its element-fitting generator is not published in the surveyed repository. No dependencies. Data CC-BY-SA-4.0 ¹.

tahze0/solar-eclipse-explorer. Python, MIT code, 0 stars, last push 2026-09-08. Reconstructs all 11,898 Canon eclipses "computed locally from NASA Five Millennium Canon Besselian elements" from eclipse_besselian_from_mysqldump2.csv, with umbra and penumbra modules, a 14-step build pipeline and reported "median 0.26% error" in path width against catalogue values on 300 reference eclipses ².

aravpanwar/besselian. Python, AGPL-3.0, 0 stars, last push 2026-09-14. Local circumstances from NASA elements intersected with OpenStreetMap populated places for 2027 August 2 in Egypt. States $k_1 = 0.272488$ for penumbral and $k_2 = 0.272281$ for umbral contacts, ΔT = 76.0 s, and that omitting ΔT in the hour angle "introduces ~32 km longitude error". Validates Sun altitude 81.69° and duration 382.5 s at greatest eclipse against published values ⁸.

Absolute-Eclipse/eclipse-data. JavaScript, MIT code and CC-BY-4.0 data, 0 stars, last push 2026-06-17. Local circumstances for 9,438 European cities for 2026 and 2027 from "Besselian elements sourced from F. Espenak (NASA/EclipseWise)", with ΔT, lunar-limb adjustments and refraction documented, elevation from Copernicus, NASA and ASTER, "accuracy within 1–5 seconds for greatest eclipse and 0.05 percentage points for partial phases" against NASA and Espenak, residuals published, and near-limit cities flagged ⁹.

johnarban/eclipse_explorer. TypeScript, GPL-3.0, 0 stars, last push 2026-07-15. A TypeScript port of NASA's JavaScript Solar Eclipse Explorer (JSEX) with SE2001.ts and SE2024.ts element files and a test file ¹⁰. ebradyjobory/eclipse-explorer. JavaScript, GPL-2.0, 16 stars, one commit in 2010. It mirrors NASA's JSEX and JLEX unchanged ¹¹.

plhery/umbra-eclipse-atlas. TypeScript, MIT, 0 stars, last push 2026-09-08. Browses 11,898 eclipses using "Besselian elements from NASA's Five Millennium Canon via the @astronomy-bundle/solar-eclipse library", Copernicus DEM skyline via Open-Meteo, and states it is "a planning aid, not an official prediction service" ¹².

gmiller123456/FiveMillenniumCanonOfSolarEclipses-Besselian-Elements. Python, 3 stars, last push 2023-03-10. Repackages NASA's elements as CSV, JS and JSON with condensing scripts. Data use requires the Espenak acknowledgment ¹³. Greg Miller's site also documents generating polynomial elements from a DE405 extract with $k = 0.2725076$ and $R_\odot = 6.957\times10^8$ m, fitting five points at $t_0 \pm 2$ h with Gauss-Jordan elimination, and matching Espenak "to approximately seven decimal places" ¹⁴, and a prediction article citing the Explanatory Supplement, Meeus, Chauvenet, Green and Smart ^{15 16}.

Frencil/eclipsetracks. Python and JavaScript, Apache-2.0, 31 stars, last push 2024-02-20. Scrapes NASA path tables into CZML for a Cesium globe and does not compute the umbra itself ¹⁷.

Geometric computation on Astronomy Engine or Skyfield

Xeratec/eclipsite. Python, MIT, 2 stars, last push 2026-08-12. Ranks observing sites by "P(clear line of sight) × eclipse depth × terrain gate × model-agreement discount × drive-time discount". Local circumstances from JPL DE421 via Skyfield, cross-checked against NASA Besselian elements to "0.36s mean difference". It adds an EU-DEM 25 m skyline via OpenTopoData and cloud ensembles. Validation gives "RMS 0.60s" against IGN and Generalitat de Catalunya durations, which "caught a solar-radius convention error worth 1.3 seconds per site" ¹⁸.

tomasrojasc/eclipse-2026. Python and TypeScript, no licence stated, 0 stars, last push 2026-08-12. Contact times for Spain via Skyfield and DE421, and beads "computed from the Moon's actual shape. The app builds a limb profile from LOLA laser altimetry for the Moon's libration at your contact time, then slides that jagged limb across the Sun's edge". It uses LRO LOLA ldem_16.img at 16 pixels per degree, about 1.9 km at the limb, with the NAIF moon_pa_de421_1900-2050.bpc kernel. There are 84 tests, including totality-band checks for 11 cities ⁵.

angristan/eclipse. TypeScript, MIT, 1 star, last push 2026-09-09. Paths by intersecting the Sun-Moon axis with the WGS84 ellipsoid using astronomy-engine's SearchGlobalSolarEclipse and SearchLocalSolarEclipse, umbra radius by similar triangles, "no database and no data files", and "Overall accuracy is a few kilometers: great for a map, not for planning an expedition to the edge" ¹⁹.

mf4633/eclipse-predictor. HTML single file, MIT, 0 stars, last push 2026-09-08. Catalogue of 1900 to 2200, about 690 events, from astronomy-engine, "Besselian elements derived from position data" to trace the path, claims "±1–2 km ground accuracy" and local contacts "to a few seconds" against NASA ²⁰.

carlok/eclipse-3d-sim. Python and React, no licence stated, 1 star, last push 2026-07-25. Astronomy Engine circumstances compared against terrain by ray-casting a WGS84 GeoTIFF DEM, optionally Copernicus 30 m from OpenTopography, plus a manual obstacle editor ²¹.

jpatka/Eclipse-Explorer. Python, MIT, 0 stars, last push 2026-08-14. Skyfield with DE421, refraction, and CSV and iCalendar output. No validation is stated, and it was developed with an LLM assistant ²².

pmisson/pyeclipsesimulator. Python, MIT, 4 stars, last push 2025-09-07. Astropy-based contacts with refraction libraries and Open-Meteo elevation, "compared with xjubier.free.fr" with discrepancies of a few seconds, "purely demonstrative", mostly LLM-generated ²³.

AstroWimSara/SolarEclipseWorkbench. Python 3.11, GPL-3.0, 20 stars, last push 2026-08-07. Camera-control workbench computing C1, C2, MAX, C3, C4, sunrise and sunset. "Astropy/IERS is used to compute Delta T (TT − UT1)", solar radius "959.95 ±0.05 arcseconds", a bundled lunar_limb_band_v1.bin, and SPICE kernels for the Moon ⁶.

sPaMFouR/SolarEclipse. Jupyter, 2 stars, 2021. It uses astropy and astroplan, is "under development", and documents no method ²⁴.

Element generators and textbook ports

i230010/sedategenerator_v2. Python, MIT, 2 stars, last push 2026-09-15. Emits eclipse dates with polynomial elements x0..x3, y0..y3, d0..d3, l10..l13, l20..l23, mu0..mu3, tan f1, tan f2, and ΔT, from a bundled ephem folder whose source is not documented ²⁵.

RyuuNeko1107/umbra-rs. Rust, Apache-2.0 or MIT provisional, 0 stars, last push 2026-09-15. "Experimental pure-Rust solar eclipse prediction engine" with VSOP87, ELP-MPP02 and IERS EOP inputs, mean limb, precision targets of ±1 to 2 s, and the statement that it is "not ready for use" ²⁶.

johanley/custom-solar-eclipse-viewer. Java, no licence stated, 2 stars, last push 2024-06-03. Local circumstances with Meeus's Elements of Solar Eclipses, 1989, as "my main reference" and the 1961 Explanatory Supplement ²⁷.

Starainrt/astro, Go, Apache-2.0, 157 stars, last push 2026-08-06; commenthol/astronomia, JavaScript, MIT, 180 stars; soniakeys/meeus, Go, MIT, 376 stars, 2019; and andrmoel/astronomy-bundle, PHP, MIT, 12 stars, are covered in open-source libraries ^{28 4 29 30}.

curiousleo/twilight, C++, 2013, integrates Sun, Earth and Moon numerically and tests eclipse geometry with tangent lines ³¹. fizyk20/eclipse-predictor, Rust, 2022, has 8 commits and no README detail ³². ArctynFox/SunSketcher, Java, 2 stars, is the NASA-partnered Android app for crowd-sourced Baily's beads imagery. Its README covers the build only ³³.

Comparison matrix

The five most complete

1. Stellarium SolarEclipseComputer.cpp: elements, limits, outlines, rise-set curves, contacts and maps with cited equations ³⁴.
2. RHerAle/Eclipse-Engine: elements fitted to DE440s, full Besselian product set, terrain skyline, radiometry ¹.
3. tahze0/solar-eclipse-explorer: full 5,000-year catalogue reconstruction from the NASA CSV with a measured path-width error ².
4. tomasrojasc/eclipse-2026: the only repository that computes Baily's beads from a LOLA DEM with a lunar orientation kernel, in its case the DE421 one ⁵.
5. Xeratec/eclipsite: local contacts from two independent methods with published residuals and a DEM skyline gate ¹⁸.

Negative findings

The thirty-repository September 2026 inventory includes no implementation deriving general northern and southern umbral limits from lunar profiles, no Watts-chart reader and no reusable general-library LOLA profile function. This scope leaves room for implementations outside the inventory and for the single-eclipse bead work listed above ⁷.

Sources compared

What a developer should do

Clone Stellarium sparsely (src/core, src/gui) and read SolarEclipseComputer.cpp end to end ³⁴. Read program.js from NASA for the minimal local-circumstance loop ³⁷. Read tomasrojasc/eclipse-2026's bead code for a worked LOLA limb construction with the correct orientation kernel ⁵. Use tahze0's pipeline as a template for catalogue-scale validation against the NASA CSV ². Avoid the LLM-generated repositories for anything but ideas, since they state so themselves ^{23 22}.

What this changes

Nothing in the stage design. The surveyed implementations do not supply a general limb-corrected path-limit library, and that LOLA-based bead prediction has a small worked precedent.

Open questions

• Obtain data/eclipses.json and the fitting script from RHerAle/Eclipse-Engine and compare its 2026 elements against NASA's page ¹.
• Obtain tomasrojasc/eclipse-2026's limb-profile module and test it against Jubier's 2026 Baily's beads diagram for one site ^{5 38}.
• Obtain SolarEclipseWorkbench's lunar_limb_band_v1.bin format description ⁶.
• Obtain the @astronomy-bundle/solar-eclipse npm package source to see which element set it bundles ¹².