Repository navigation
Conversation
@types/node 17 ships no node:test typings, so the upcoming test suite cannot be type-checked without this bump. @types/node 26 only declares typesVersions for TS <= 5.7, so TS 4.5 would fall through to the root typings and fail to parse them. Bump both. The @tsconfig/node16 base is deliberately left in place so the emitted JS shape is unchanged for consumers. The consequence is that the type surface now describes a far newer Node than the compile target, making "node16 support" advisory rather than enforced. Noted in AGENTS.md. Verified: npm run build is clean under TS 5.9.3 / @types/node 26.6.3 with no new errors in lib/*.ts.
-dirtecting the config store used to require -D, which persists. That made
it awkward to run mourt against a scratch data dir without rewriting the
real config, which is an easy way to lose settings.
dir and plugin.taskw.command stay file-backed via configstore's dotted
lookup, so a single knob is enough: point at a different config file.
Precedence is -c/--config > $MOURT_CONFIG > configstore's default.
`-c <path>` alone prints the resolved path, as does `-P path`, so the
active config is never a guess.
The store moves from module scope into the Cli constructor, since argv has
to be parsed before the store is opened. That also gives tests a place to
inject a throwaway store.
Two configstore sharp edges are handled explicitly:
- Its constructor writes the file when given a truthy defaults argument,
so createConfig passes none. Passing one would rewrite the config on
every read.
- Its `all` getter silently empties any file it cannot JSON.parse, so
resolveConfigPath validates the target first and throws instead. A typo
in --config must not blank a config file.
bin/mourt now wraps construction in try/catch: a bad --config throws
outside the run() promise chain, so it would otherwise escape as a raw
stack trace rather than a clean message.
Locals in config.ts are prefixed (nodefs/nodepath/openConfig) because
lib/*.ts are CommonJS scripts sharing one global scope. Using `export {}`
instead would emit an __esModule marker into the CJS and could confuse
bundler interop.
Verified: redirected stores write only to the target; a corrupt --config is
rejected with the file left byte-identical; the real configstore file hashes
the same before and after.
59 tests across 7 files, run with no new dependencies: npm test # npm run build && tsc -p tsconfig.test.json && node --test tsconfig.test.json compiles to dist-test/ rather than dist/, so test code can never reach the published artifact. This matters because the files field is dist/**/*.js: adding test/ to the main tsconfig would have shipped the tests in the tarball. Coverage is weighted toward the behaviour that was previously only verified by hand: - getList()'s four-case project filter, including the documented rule that entries with no taskwarrior field are excluded under a project filter - plugin-taskw via a real spawned fake `task` script, covering the empty array, non-array JSON, oversized-field truncation, and the fact that taskwarrior's stderr reaches the rejection message rather than a bare "non zero exit". A fake executable is used instead of an injected function on purpose: the stderr fix only manifests at the process boundary. - config resolution precedence, rejection of unparseable JSON, and the fact that opening a store does not rewrite the file - doMessage's ISO date agreeing with both its filename and its local YYYY/MM/DD directory. No clock is injected, so assertions check internal consistency rather than a fixed timestamp. - e2e runs spawning bin/mourt, which is what would have caught the dist/ layout break that tsc allowed silently. Verified by mutation: reverting the requireBaseDir guard, the j[0] guard, the config precedence order, and the bin/mourt require path each turn the suite red (8, 2, 4 and 8 failures respectively). All were restored and lib/ matches HEAD. Every test is scoped to a temp config and data dir. The suite hashes the real configstore file before and after and it is byte-identical, so running the tests cannot rewrite the user's real settings. No real journal content is used anywhere in the suite; values are synthetic. The AGENTS.md "Handling real data" section records why.
The matrix was pinned to 10.x/12.x/14.x, all long EOL. Two of those predate node:test entirely, so the new suite could not have run there. matrix is now 22.x (Jod), 24.x (Krypton) and 26.x (Current). 26.x is not LTS yet but reaches it in Oct 2026, and testing it now surfaces breakage before promotion day rather than after. Also updated: - actions/checkout v2 -> v7, actions/setup-node v1 -> v7, and enable `cache: npm`. - `npm run build --if-present` is gone; `npm test` runs `npm run build` first, so the build is covered without a redundant step. - `npm test || echo 'HEY! write some tests, already'` is now a plain `npm test`. That guard swallowed every failure, which is indefensible now that the suite is real. A failure must fail the build. - fail-fast: false, so all three versions report instead of one cancelling the rest. engines.node is set to >=22.3, the floor implied by the tests: node:test glob arguments landed in Node 21, and t.after() in 22.3. Without the engines field nothing warns on an unsupported runtime. --test-reporter=spec is now explicit. Node 22 defaults to tap and Node 24+ to spec, so CI logs were inconsistent between matrix entries. Verified locally against real interpreters rather than assumed: v22.23.3 and v24.21.0 downloaded from nodejs.org, full build plus 59/59 tests green on both, and again on the local v26.7.0. The YAML parses. The user's real configstore file is byte-identical after the runs.
The list had drifted out of date in ways that would mislead:
- "No tests exist yet / npm test exits 1" was still asserted in the
Commands section and Gotchas, months after that stopped being true. Now
states the real 59 cases and the build-then-test ordering.
- Test suite, config redirect and CI are recorded as done under a new
"Later" section instead of being missing entirely.
Four genuinely open items are now listed rather than implied:
- CI is Linux-only, so macOS and Windows are untested. The binary and
configstore both have platform-specific behaviour.
- The semver@6.3.0 ReDoS advisory, and what it would actually take to
clear it.
- No lint or typecheck-only script exists.
- `getList()` remains O(all entries).
The two `// TODO` markers still in lib/ are recorded as decisions that
need making, not as chores.
The one place that kept a historical note ("npm test failing meant no
tests yet, not tests are broken") is reworded so it describes what was
true when written rather than claiming to be the current state.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.