fix(builder): the Linux builder shows its clock and countdown during a build #2150

Merged

jarynclouatre merged 1 commit from fix/linux-builder-clock into main

2026-11-11 20:38:06 +00:00

jarynclouatre commented

2026-11-11 05:05:08 +00:00

(Migrated from github.com)

During a build, the line under the progress bar in the Linux builder stays empty: no elapsed time, no compile clock, no "about N minutes left". Only an update shows it.

tick writes that line and calls itself again a second later only while make_game.py runs. build reached it through show_progress, which runs before make_game.py is started, so the first call found no process and the clock never started. start_make_update calls tick again once make_game.py runs, which is why updates were fine. It has been this way since the Linux builder was added, so the countdown added later has never shown in a Linux build.

build now starts the clock after starting make_game.py, as start_make_update does, and show_progress no longer calls tick, since nothing runs yet at either of its callers. on_exit clears the line: with the clock running, a build cancelled or stopped while compiling would otherwise keep "about 20 minutes left" under "The build stopped". A finished build already says how long it took in the line above.

Testing (Linux, Tk 8.6)

The real window on Xvfb, with make_game.py replaced by a stand-in that prints @@stage compile and @@eta 600 and then sleeps. Before, the line is empty for the whole run. After, it reads Elapsed 0:00:01 · compiling 0:00 · about 10 minutes left and counts each second, and it's empty again after Cancel. An update counts as before.
tests/test_linux_builder_clock.py (new) runs the builder's own tick and on_exit on a stand-in window, without Tk or a display. tests/test_linux_builder_jobs.py now expects the clock to start after the process. On main, 1 and 7 of them fail.
Not tested: a full build from the AppImage.

During a build, the line under the progress bar in the Linux builder stays empty: no elapsed time, no compile clock, no "about N minutes left". Only an update shows it. `tick` writes that line and calls itself again a second later only while make_game.py runs. `build` reached it through `show_progress`, which runs before make_game.py is started, so the first call found no process and the clock never started. `start_make_update` calls `tick` again once make_game.py runs, which is why updates were fine. It has been this way since the Linux builder was added, so the countdown added later has never shown in a Linux build. `build` now starts the clock after starting make_game.py, as `start_make_update` does, and `show_progress` no longer calls `tick`, since nothing runs yet at either of its callers. `on_exit` clears the line: with the clock running, a build cancelled or stopped while compiling would otherwise keep "about 20 minutes left" under "The build stopped". A finished build already says how long it took in the line above. **Testing (Linux, Tk 8.6)** - The real window on Xvfb, with make_game.py replaced by a stand-in that prints `@@stage compile` and `@@eta 600` and then sleeps. Before, the line is empty for the whole run. After, it reads `Elapsed 0:00:01 · compiling 0:00 · about 10 minutes left` and counts each second, and it's empty again after Cancel. An update counts as before. - `tests/test_linux_builder_clock.py` (new) runs the builder's own `tick` and `on_exit` on a stand-in window, without Tk or a display. `tests/test_linux_builder_jobs.py` now expects the clock to start after the process. On main, 1 and 7 of them fail. - Not tested: a full build from the AppImage.

Joey-Magnus (Migrated from github.com) approved these changes

2026-11-11 20:06:54 +00:00

Joey-Magnus (Migrated from github.com) left a comment

Nice catch, and a clean fix. tick only reschedules itself while self.proc is running, so calling it from show_progress before the Popen could never start the clock. Moving the call to after the process starts matches start_make_update. The show_progress call in begin_update loses nothing either, since no process runs there yet. Clearing time_text in on_exit is also needed now that the clock really runs. The pending after(1000) tick then finds proc as None and stops, which your test_it_ends_with_the_build covers. show_progress already resets eta and compile_started, so a second build after a stopped one doesn't show the first build's countdown.

I ran the new test_linux_builder_clock.py, the updated test_linux_builder_jobs.py and the other checks that read builder/linux/hookshot_builder.py (test_builder_*, test_desktop_exec, test_disk_warning, test_prepare_at_build, test_make_game_ngx, test_linux_play_diagnostics) at the PR's head on Linux: all pass, except two .NET-only checks that skip here. CI is green, and the branch still merges cleanly onto current main.

Nothing to change from my side.

Nice catch, and a clean fix. `tick` only reschedules itself while `self.proc` is running, so calling it from `show_progress` before the `Popen` could never start the clock. Moving the call to after the process starts matches `start_make_update`. The `show_progress` call in `begin_update` loses nothing either, since no process runs there yet. Clearing `time_text` in `on_exit` is also needed now that the clock really runs. The pending `after(1000)` tick then finds `proc` as `None` and stops, which your `test_it_ends_with_the_build` covers. `show_progress` already resets `eta` and `compile_started`, so a second build after a stopped one doesn't show the first build's countdown. I ran the new `test_linux_builder_clock.py`, the updated `test_linux_builder_jobs.py` and the other checks that read `builder/linux/hookshot_builder.py` (`test_builder_*`, `test_desktop_exec`, `test_disk_warning`, `test_prepare_at_build`, `test_make_game_ngx`, `test_linux_play_diagnostics`) at the PR's head on Linux: all pass, except two .NET-only checks that skip here. CI is green, and the branch still merges cleanly onto current main. Nothing to change from my side.