Skip to content

Retire the Sciweavers renderer, keep its de-renderer - #72

Merged
Divide-By-0 merged 1 commit into
masterfrom
codex/retire-sciweavers-renderer
Sep 19, 2026
Merged

Divide-By-0 merged 1 commit into
masterfrom
codex/retire-sciweavers-renderer

Conversation

@Divide-By-0

Copy link
Copy Markdown
Owner

Sciweavers is retired as a renderer and kept as a de-renderer. The 404 the PR #71 probe reported is permanent, and it is not the same thing as the site being down.

Human intent

Rephrased from the request: PR #71's live probe reported "Sciweavers HTTP 404", but sciweavers.org clearly still loads — go find out what is actually broken and fix it, rather than leaving a renderer that reports a failure.

What was actually checked

Both things are true: the site is up, and the endpoint AutoLatex uses is gone.

Every Sciweavers entry in Common.getRenderer is built on tex2img.php, a GET URL that returns the image directly. That file now returns the site's own 404 page. Checked 2026-09-18:

Variation tried Result
https://www.sciweavers.org/tex2img.php?bc=…&eq=… (our URL) 404
…?eq=…&bc=… (the order sciweavers' own editor emits) 404
http:// (301s to https://), apex sciweavers.org 404
curl, Chrome, Googlebot and Apps-Script-shaped user agents 404
With a session cookie and a Referer from the editor page 404
/tex2img.php with no query string at all 404
Real Chrome, screenshot below 404

Meanwhile https://www.sciweavers.org/free-online-latex-equation-editor returns 200 and its Tex2Img editor still renders. Its working path is now a POST to /process_form_tex2img that writes a one-off image to /upload/Tex2Img_<id>/render.png. Two identical submits 3 seconds apart returned Tex2Img_1789770006 and Tex2Img_1789770009 — the id is per-request and carries none of the LaTeX. AutoLatex de-renders by reading the source back out of the image URL, so that URL shape cannot replace tex2img.php even though it produces a correct image. There is no working GET endpoint to point at.

Their editor still hands out a tex2img.php link as the embed URL after a successful render. That link 404s too, which is probably why this went unnoticed on their end.

sciweavers.org is up the endpoint we use is gone
editor loads tex2img.php 404

What changed

  • Common.retiredRendererFamilies lists Sciweavers and Sciweavers_old, and getRendererOrder skips them. The render order goes from [1,2,3,4,5,6,7,8] to [1,2,3,4,6], dropping three guaranteed-failing renderer attempts — two fetches and a 50 ms sleep each — from every equation that gets past Codecogs and Texrendr.
  • The Sciweavers option is gone from the Docs, Slides and Sheets renderer selectors; it failed 100% of the time.
  • A saved sciweavers preference migrates to Automatic in normalizeRendererPreference (so savePrefs writes the dead value away on the next render) and in each Sidebar.ts, alongside the existing Codecogs mapping — otherwise the selector renders blank and posts a dead choice back.
  • Docs' post-MathJax fallback list drops the two Sciweavers families it could never reach.
  • De-rendering is untouched. Indices 5, 7, 8 and 11 stay in getRenderer and within capableDerenderers, so equations rendered before the shutdown still round-trip.

Validation

  • npm test: 133 passing, up from 128. The new tests/retired-renderers.test.js pins both halves — reverting the render-order filter fails 3 of its 5 tests, while the de-render and sidebar tests stay green, which is the point.
  • npm run test:render-browser: 40 browser settings, 256 real PNG renders, pass.
  • npm run test:render-services: 40 server settings, 8 distinct live requests, 0 unavailable, exit 0. On master the same probe made 12 requests and 4 failed. Codecogs and Texrendr are unchanged; the four dead requests are simply no longer made.

Live external renderer responses

Two hard-coded values that would have silently absorbed this were derived instead: the matrix test's cases.length (100 → 80) and the live probe's gallery row list, which would have crashed on a renderer that no longer has results.

Full write-up and the probe's results.json: tests/evidence/sciweavers-retired. tests/README.md and PR #71's evidence README now point at it instead of leaving the 404 reading as a transient outage.

Release status

Code only — no Apps Script push, no version, no Marketplace change. Docs remains v95 and Slides v15.

🤖 Generated with Claude Code

sciweavers.org retired tex2img.php, the GET image endpoint all four Sciweavers
renderer entries are built on. The site itself is up and its Tex2Img editor
still renders, which is why the 404 read like a blip: verified 2026-09-18 that
tex2img.php returns the site's own 404 page over http and https, on www and the
apex domain, in both parameter orders, with and without a session cookie and
Referer, under curl/Chrome/Googlebot user agents, and in real Chrome. The site's
working path is now a POST to /process_form_tex2img returning a one-off
/upload/Tex2Img_<id>/render.png whose id changes per submit and carries none of
the LaTeX, so it cannot satisfy the contract that an equation's image URL is
de-renderable back into source.

Rendering therefore skips the retired families, which removes three
guaranteed-failing renderer attempts (two fetches and a sleep each) from every
equation that gets past Codecogs and Texrendr, and removes a sidebar option that
failed 100% of the time. A saved "sciweavers" preference migrates to Automatic
instead of failing every equation.

Sciweavers stays in getRenderer and within capableDerenderers so equations
rendered before the shutdown keep de-rendering; tests/retired-renderers.test.js
pins both halves so neither can be undone by checking that the homepage loads.

The live probe now reports 40 server settings, 8 distinct requests, 0
unavailable (was 12 requests, 4 failing). Evidence in
tests/evidence/sciweavers-retired/.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Divide-By-0
Divide-By-0 merged commit f9288ba into master Sep 19, 2026
3 checks passed
@Divide-By-0
Divide-By-0 deleted the codex/retire-sciweavers-renderer branch September 19, 2026 02:36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant