chore: one round per platform instead of checkup - #51
Merged
Conversation
`checkup` was lint, typecheck, unit and e2e behind one name, and it said nothing about what a platform can actually run. Three rounds say it: test:all:mac everything here, ending with the hardware specs test:all:windows everything here, and socat is not here test:all:linux everything, plus the two that wait for a person Each puts the unattended work first, so the only thing left when it stops is the part that needs someone at the keyboard. The name a contributor runs before pushing is `verify`, with exactly what `checkup` had. The rounds are for cutting a release: they package, they take screenshots, and two of them wait for hardware.
assertOpenable built its device paths with path.join, which is the host's join. The string it builds is a Linux device quoted back to the reader, not a path on the machine running the check, so on Windows the unit suite got \dev\ttyACM0 and failed. The join in unreadablePorts stays as it is: that one does touch the running machine, and there it should follow the platform. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014HBuLkUjywRUCPxQSuSUSB
Closing a real COM port took about 4.5s on the Windows VM these specs run on, against a 5s wait in the test. The whole hardware suite lost that race: disconnectClient failed outright, and the two disconnects in the reconnect spec would have failed the same way had maxFailures let the run get there. Measured with a temporary timer: the disconnect completed at 5314ms with only "Disconnected from server" in the snackbars, so close() called back on its own and the app's own 5s watchdog never fired. Slow, not stuck. The watchdog in modbusClient.disconnect still sits at 5000ms, right on top of an operation that takes ~4.5s here. Left alone: that is app behaviour, and it may well be this VM's USB passthrough rather than the app. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014HBuLkUjywRUCPxQSuSUSB
Both hardware specs stopped at a page.pause() for someone to pick the COM port by hand. That is why the suite sat outside every round on this platform: an unattended run waited forever, and a release round could not include it. The board is identified by USB vendor id, read from the same serialport the app uses. Not by manufacturer: on Windows that reads "Microsoft", because the generic usbser driver claims the device, while vendorId still says 2341. Only the two genuine Arduino ids count. A clone presents the id of whatever bridge it carries, and those chips sit on hundreds of unrelated adapters, so matching them would let the suite read registers off a random dongle. The failure prints every port it saw, so extending the list is obvious. Two boards fails rather than guesses. No board skips the suite, so a machine without hardware does not block a release round. test:all:windows now ends with the hardware specs, like the other two rounds. Measured: 32 passed in 1.6m, no keyboard. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014HBuLkUjywRUCPxQSuSUSB
The hardware specs paused for a flat 5s or 10s after clicking read and then asserted. Both were guesses, and one of them was wrong: the illegal-address test timed out at 30s on Linux, four runs out of four. Two things were hiding in there. The read that follows enabling read-configuration is dropped when it arrives too soon after the toggle. setReadConfiguration waits for Mui-selected, but that class lands before the app can serve a read. Measured on the Arduino: 10ms before the click still fails, 100ms passes. READ_SETTLE_MS gives it 250ms. This works around the race rather than fixing it -- clicking Read by hand within that window loses the read too. And two assertions proved nothing. Both this suite and connectAndRead checked rowCount > 0, which stays true when the read fails completely, because the grid keeps the rows it built from the config. They now wait for an actual cell value, which is both stricter and faster. The suite drops from 1.5m to under a minute: the illegal-address test from a 30s timeout to 2.5s, the re-read from 5.5s to 1s. The 2000ms after connect stays. That one is not a guess -- the board reboots on DTR when the port opens.
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
1 participant
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.
checkupwas lint, typecheck, unit and e2e behind one name, and it said nothing about what a platform can actually run. Three rounds say it:test:all:mactest:all:windowstest:all:linuxEach puts the unattended work first, so the only thing left when it stops is the part that needs someone at the keyboard.
checkupwas also the command CONTRIBUTING told a contributor to run before pushing, five times over. The rounds are far too heavy for that: they package, they take screenshots, two of them wait for hardware. So the pre-push name isverify, with exactly whatcheckuphad, and CONTRIBUTING points at it.The command table there grew the four suites that were missing from it as well:
scan-perf,privileged-port, and the three rounds.