Repository navigation
Conversation
system_sleep_changed() discarded the cached native projection on both sleep edges, so the resume warmup repeated the full hardware inventory (its own AKS/ACM authorization, Bridge inventory and Catacomb restore) before it could prepare the verifier. Both take the machine-wide operation lock without blocking, so they could only run in sequence. The sleep edge now only advances the inventory generation, so a read straddling it cannot publish, and keeps a completed native projection. runtime_projection() still rechecks the authority and Catacomb/journal state tokens before reuse, native matching still repeats live SEP inventory reconciliation and attestation, and verifier preparation is still released on sleep. On the MacBookPro16,1 reference machine one lid-close resume reached the lock-screen fingerprint prompt 2.42 s after thaw, versus 4.34-4.52 s for the three preceding resumes on the old code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner
|
Thanks for this, and for measuring it: on your MacBookPro16,1, the time from resume to the prompt went from about 4.4 s (three resumes before) to 2.4 s (one resume after). We're closing it as superseded, because Resume time matters there too, and we've recorded your result and evaluation for that work, with credit. When that path is ready to test on a 16,1, we'd like to ask you to measure it. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What changed
T2Backend.system_sleep_changed()used to callinvalidate_inventory("system_sleep")on both sleep edges. That dropped the cached native presentation projection, so the resume warmup (already started on logindPrepareForSleep(false)) had to repeat the full hardware inventory before it could prepare the verifier. That inventory runs its own AKS/ACM authorization, Bridge inventory and Catacomb restore. The inventory collector and the match worker both take/run/t2-touchid/operation.lockwithLOCK_NBand both open AKS/ACM, so the two steps can only run one after the other. On the reference machine that cost about 2 s of the time from resume to the lock-screen prompt.The sleep edge now only advances
inventory_generation. A read that straddles the edge still cannot publish, and a completed native projection is kept. Nothing else changes:runtime_projection()still checks the account-authority token and the committed Catacomb / mutation-journal state token before any reuse.This matches the existing contract in
docs/FPRINT_INTEGRATION.md§1: presentation metadata may stay cached while authority, Catacomb and journal metadata are unchanged. That section now also says explicitly how sleep is handled.Measurements
MacBookPro16,1, Linux
7.2.11-arch1-Watanare-T2-1-t2-amdoff, deep sleep started by a lid close, timed fromuser.slicethaw. Full table indocs/evaluations/touchid-resume-presentation-cache-2026-10-27.md.The verifier's AKS phases did not change (keybag retain, ACM context authorization, identity unlock). There is only one candidate resume, so this is not a latency distribution.
Review notes
VerifyStartinstead of atListEnrolledFingers. Both fail closed, and password fallback is unchanged.src/t2-fprintd.pyand restart fprintd.docs/SLEEP_POLICY.md.Tests
tests/test_t2_fprintd.py:main.python -m unittest discover -s tests, 1654 tests OK (1 skipped).tools/privacy-check.shandenrollment_research/scripts/check-public-tree.shpass.🤖 Generated with Claude Code