You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
<!-- agent-pmo:74cf183 -->
## TLDR
Adds plain-`http://` support to the VS Code OpenAPI importer, renders
saved playlist reports in an in-editor webview (no more blocking
external-app modal), makes the redirect e2e test hermetic, upgrades the
extension dev toolchain, adds website analytics + a Deslop gate, and
fixes two `make ci`/CI-divergence bugs.
## Details
**OpenAPI download over HTTP(S)** —
`src/Napper.VsCode/src/openApiDownloader.ts`
- `downloadSpec` selects the `http` client when the URL begins with
`http://`, else `https`, so self-hosted/internal specs served over plain
HTTP can be imported. Redirect-following and error handling are
preserved; the response handler is shared between both clients.
- New named constant `HTTP_SCHEME_PREFIX = 'http://'` in `constants.ts`.
**Playlist report now opens inside VS Code** —
`src/Napper.VsCode/src/extension.ts`
- `savePlaylistReport` previously called
`vscode.env.openExternal(Uri.file(report.html))`. On any host without an
OS file→app association (headless CI, the extension-test window) that
raises a blocking **"An error occurred opening an external program — No
application found to open URL"** modal that must be dismissed by hand
and stalls the automated run.
- The report HTML is fully self-contained (inline styles), so it now
renders in an in-process webview panel (`REPORT_PANEL_VIEW_TYPE`) titled
`<playlist>-report`. The file is still written to disk and the "Report
saved" message still shows. Hermetic, better UX, no external app needed.
**Dev-dependency upgrades** — `package.json` + regenerated
`package-lock.json`
- `@types/node` 25.3→25.9, `@types/vscode` 1.95→1.120, `@vscode/vsce`
3.7→3.9, `eslint` 10.0→10.4, `mocha` 11.7.5→11.7.6, `prettier`
3.8.1→3.8.3, `ts-loader` 9.5→9.6, `typescript-eslint` 8.56→8.60,
`webpack` 5.105→5.107, `webpack-cli` 6→7 (major). Marketplace `preview`
flag flipped `true`→`false`.
**Website Google Analytics** — `website/eleventy.config.js`
- New Eleventy transform injects the gtag.js snippet (id `G-JNZW4P4DYD`)
before `</head>` on every `.html` page.
**CI-infrastructure fixes** — `Makefile`
- `_test_vsix` now runs `npm run compile:e2e` (alongside `compile` +
`compile:tests`). Previously `make` only compiled the unit suite, so the
e2e suite ran from a **stale** build — masking source changes (it ran
the old httpbin-based redirect test, which fails when httpbin is down).
CI was unaffected because it runs `npm run pretest` (which includes
`compile:e2e`); this realigns `make ci` with CI.
- `_coverage_check` parses the Rust cobertura `line-rate` with `xmllint
--xpath` instead of a greedy `sed` regex. The report is a single line,
so the old regex matched the **last** `line-rate` (the empty `src/tests`
package = `0`) and reported 0%, aborting the build despite genuine 100%
line coverage. Uses a real XML parser per the repo's "no regex on
structured data" rule.
**Repo standards / tooling**
- `.deslop.toml` (new): duplication gate `max_duplication_percent =
15.2` over the Rust Zed extension (Deslop does not yet analyse F#);
ratchet-down only. Implements REPO-STANDARDS-SPEC `[CI-DESLOP]`.
- `.gitignore`: stop ignoring `.idea/` and `.vscode/` so shared IDE
config is committed per `[GITIGNORE-RULES]`.
- `.claude/settings.json` (new): `autoMemoryEnabled: false`.
## How Do The Automated Tests Prove It Works?
- **Redirect (hermetic)** — `openApiImport.e2e.test.ts › downloadSpec
follows redirects` spins up a local `http.createServer` that
302-redirects `/redirect` → `/spec.json`, then asserts `downloadSpec`
follows the redirect over plain HTTP and the parsed body has an
`openapi` field. Exercises the new `http://` path with zero external
dependency. Verified: `vscode-test --grep "follows redirects"` →
passing.
- **Report opens in-editor** — `playlist.e2e.test.ts › save report
command creates HTML report file after playlist completes` now
additionally asserts (through the UI) that a webview tab labelled
`smoke-report` appears after the Save Report command — proving the
report renders in VS Code with no blocking external-app modal. The csx
save-report tests also pass. Verified: `vscode-test --grep "save
report|follows redirects"` → 4 passing, no modal.
- **Full `make ci`** (lint → test → build) passes locally end-to-end:
Fantomas/Prettier/cargo-fmt, `dotnet build -warnaserror`, ESLint,
clippy; F# tests (Napper.Core 97.9%, DotHttp 90.5%, Napper.Lsp 100%),
Rust tarpaulin (100%, now parsed correctly), TS unit + extension-host
e2e suites, all coverage thresholds, the universal VSIX package, and the
NativeAOT CLI publish + `napper --version` contract check.
## Breaking Changes
None. The HTTP(S) download change is additive (HTTPS stays default); the
report now opens in-editor instead of an external browser (no behavioral
loss — the file is still saved); dependency bumps are dev-only; Makefile
fixes only touch build orchestration. All verified green by `make ci`.
## Follow-up: de-flake the e2e suite (drop httpbin.org)
The first CI run on this PR went red on a pre-existing, unrelated flaky
test — `runfile.e2e › running via URI opens same response panel as via
document` — which ran a fixture against **httpbin.org** and timed out
(30s) when httpbin was slow. httpbin.org is chronically unreliable (the
redirect test was already migrated off it). To make CI reliably green:
- Renamed fixture `get-httpbin.nap` → `get-shorthand.nap` and pointed it
at a reliable real API (`jsonplaceholder.typicode.com/todos/1`). Updated
all six e2e files that referenced it; the copy-curl URL assertion and
the run-file test name were updated to match.
- `openApiImport.e2e` `NONEXISTENT_URL` now uses a reliable
jsonplaceholder 404 path instead of `httpbin.org/status/404`.
- The only remaining httpbin strings are in two explorer tests that
never make a network call (they assert file/tree state) plus one
explanatory comment.
## Follow-up 2: correct the Rust coverage threshold (CI run 2 & 3)
With the parser fixed, CI surfaced two more environment gaps:
- **`xmllint` is not installed on the GitHub ubuntu runner** (exit 127).
Switched the cobertura `line-rate` parse to `python3` + stdlib
ElementTree — a real XML parser pre-installed on both GitHub runners and
macOS.
- **Phantom Rust threshold.** tarpaulin's line attribution is
platform-divergent: macOS reports 67/67 = 100%, Linux CI reports 66/68 =
97.1% for identical code (the two "uncovered" lines — `match name {` and
a struct-literal field — are already exercised by existing tests). The
99% threshold had been auto-ratcheted from the inflated macOS number, so
CI could never meet it. Fixed by pinning the Rust threshold to 96
(`floor(CI%)−1`, matching every other project) and making the Rust
coverage check **check-only** — auto-ratcheting platform-divergent
coverage captures the wrong (highest) measurement. F#/TS ratchets are
unchanged.
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
![dependabot[bot]](https://avatars.githubusercontent.com/in/29110?v=4&size=40){kind=avatar}
0 commit comments