Skip to content

feat(ui): update the built-in MatGUI to 6.15.0 - #174

Open
MathiasVDA wants to merge 9 commits into
mainfrom
claude/project-thread-uprney
Open

MathiasVDA wants to merge 9 commits into
mainfrom
claude/project-thread-uprney

Conversation

@MathiasVDA

@MathiasVDA MathiasVDA commented Nov 22, 2026 •

Copy link
Copy Markdown
Contributor

What and why

MatGUI 6.15.0 bundles yasgui-explain-plugin 1.1.0 (Matdata-eu/yasgui-explain-plugin#9), whose Explain view knows TripleTrail: it recognises the server by the Server: TripleTrail/<version> header (#67) and draws the plan from /{ds}/explain. This PR vendors that build and removes the interim TripleTrail engine that #67 added to the Query page's app.js, since the bundle now does the same.

  • ui/vendor/matgui/: yasgui.min.js and yasgui.min.css from the npm package @matdata/yasgui@6.15.0, with the source map comments removed and gzipped (gzip -9 -n), as before. The fonts and images are byte-identical to 6.13.0.
  • app.js: the Explain subclass, tripletrailEngine and explainPlanOf are removed. MatGUI loads as before.
  • 6.13.0 → 6.15.0 in NOTICE, the vendor LICENSE header, ui.rs, the --matgui-url help, docs/server.md and the CDN test.
  • Other user-facing changes from MatGUI 6.14/6.15: a map click lists every feature under the pointer; the query editor can be hidden, and share links can open that way. They are noted in the CHANGELOG.

The branch also holds a no-op merge (-s ours) of the already-merged #67 branch history, so I didn't have to force-push. The diff against main is only this change.

How it was tested

  • tests/admin.rs::the_query_page_explain_view_reads_the_plan_of_this_server now checks that the served MatGUI carries the plugin's TripleTrail engine and header detection, and that /{ds}/explain answers the form request the plugin sends with the fields it reads. Also checked: the Server header on /explain and /sparql. 37 admin tests pass.
  • Chromium (Playwright) against a local server with the new build: run a query, open Explain, and you get "TripleTrail · 5 operations" with the plan tree, rows and ms, and the algebra. The app.js override is gone.
  • cargo fmt --check, cargo clippy -p tripletrail-server --tests -- -D warnings, node --check ui/app.js.

Documentation

  • No user-facing change, or:
  • The user guide (docs/guide/) shows it: "See how a query ran" in 02-first-steps.md stays true unchanged
  • The reference page in docs/ (server.md: Web UI, --cors, MatGUI version) and the CHANGELOG entry describe it

MatGUI's Explain view only knows QLever, GraphDB, Stardog, Blazegraph and
Neptune, so on the Query page it said "No engine recognised for this
endpoint" and showed no plan. The Query page now registers a TripleTrail
engine with it: for this server's own endpoints it POSTs the query to
…/explain (with the admin token and view-as headers) and draws plan.plan
as a tree with the rows and time of each step.
a_saved_graph_is_read_again_by_a_new_engine removed its folder with
remove_dir_all while the engine's background save of the rebuilt graph
could still write into it, which failed on CI with "Directory not empty"
and left the folder behind. It now uses tripletrail_testdir::tempdir,
removed after the engine is dropped and retried while a writer finishes.
Answers said "Server: tiny-http (Rust)". They now say
"TripleTrail/<version>", and --cors exposes the header, so the Explain
plugin of MatGUI and YASGUI can recognise the engine (the plugin's
TripleTrail engine keys on it) on any host, not only the Query page.
The Explain plugin of MatGUI and YASGUI does not recognise TripleTrail
yet; its TripleTrail engine waits for a plugin release.
MatGUI 6.15.0 bundles yasgui-explain-plugin 1.1.0, whose Explain view
knows TripleTrail: it recognises the server by its Server header and
draws the plan from /{ds}/explain. The Query page's own TripleTrail
engine for that view is no longer needed and is removed. The build files
are the npm package's, with the source map comments removed, gzipped.
The map cancels tiles it no longer needs as it zooms to the data, and
Playwright cannot read the body of such a response: the wait for a tile
with features then failed with "No data found for resource" (once in
four local runs, and on CI). A tile whose body cannot be read now just
does not count.
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