[bettterlabs]

What and why

The start after a cold start spends minutes in the prepare step, and the log says no base was compiled: here 761.61 s on an M4, the same pattern as #1747 (309 s, 0 compiled).

Cause. The timer's flush (10 s after a burst of new bases) runs during a prewarm that doesn't compact, so a cold start writes a run file every 10 s: 80 files here. More than PSO_ARCH_COMPACT run files make the next start a compaction. Its prewarm captures every base it takes from those files again from the original shader functions (the #342 workaround), a full compile each. A capture counts neither as compiled nor apart from the lookups, so the prepare line reads 0 compiled.

Change:

• Timer flushes wait: arch_flush_from holds the timer's flushes for any prewarm, as it already did for a compacting one. A prewarm then writes a file per ARCH_PERIOD_MAX. Full periods and failed captures are still handed over at once.
• Counter: the prepare line says how many bases were captured again, and pipeline bases so far also their time.

Trade-offs:

• A cold start that ends in its prewarm leaves about a period of captures unwritten, built again at the next start, where it used to leave up to 10 s of them.
• More than 16,000 bases: a cold start still writes more than 8 full files, and the next start compacts them. That's derived from the code, not seen.

One of two independent PRs; the other, #1880, compaction by size tier, deals with the files play adds afterwards.

Checked

• Tests (M4, macOS 27.2, on 270b6fd):

  • all 124 gfx tests pass.
  • t_pso_archive: 172 checks (162 on main). test_pso_archive.py: 3 tests (1 on main).
  • With the old flush rule, 8 checks and both new fresh-process tests fail; without the counter, 1 check fails.

• Game run: Ocarina of Time 0400486024DEC000 (v131072). The host was main bd979ce + eed7c42, with and without this change; this Mac's shader library is that chain, and the archive code there is the same as main's. Each start used a copy of the device archive's prebuilt files with a 30,838-line prewarm list, stopped at the title, one run each with nothing else running:

  	Cold start	Files written	Next start
  before	861.94 s	80	761.61 s, compacting all of them (8,859 bases given to the new files)
  with the change	716.20 s	5	3.30 s, no compaction

• Device: Mac mini M4, 16 GB, macOS 27.2.

• Not checked: a build of current main in a game; iOS; lint-c; regress.py check (other games named by impact: Breath of the Wild, Pokémon Scarlet, Tears of the Kingdom).

Docs

• The finding is in its game note or its docs/reference/<subsystem>.md entry, and python3 dm/tools/docs_symptoms.py was run (the symptom index)
• No game files, keys or data extracted from a game in the change or the attachments

[seedds]

Landed as 4a81561, with you credited as co-author. Thanks for the clear diagnosis and the measurements.

Your change went in as it was: a prewarm's timer flushes wait for a full period, and the compaction counter. The code comment now says 80 files, as your docs do. On top of it, what the archive holds unwritten is now written when the player quits (atexit for ⌘Q, closing the window and exit(), waiting up to 10 s; the graceful exit within its 1.5 s). It also starts writing when the app goes to the background or the window is minimized. So a quit during a cold start's prewarm no longer loses up to a period (2,000 bases). A write the exit cuts short leaves no compacting marker, which would otherwise turn compaction off on that system build. A crashed run's marker stays. test_pso_archive.py has a fourth test for the quit.

#1747 stays open: the files written during play are #1880's part, and the reporter's retest is still to come.