Skip to content

Commit ef8fc5d

Browse files
luisleo526claude
andcommitted
CI D1 fix 2: the Opus read of the D1 commit (kernel registration floor, docs that still describe the old preflight wiring, AGENTS.md)
Findings of the read of 11cd290 applied on top of 1b01244: - scripts/ci_verify.py: EXCLUDED_REGISTERED_MIN['kernel'] is 337 (KERNEL_MIN_TESTS), not 309. The 371 - 62 = 309 offset left the PR kernel leg 17 rows below the kernel registry at the vintage of the other keys (326) and let it lose 62 registrations unnoticed, against 34 before D1; 337 keeps today's 34-row slack and stays below the 371 registered in all 16 sampled pull-request and push runs. 'release' stays 895. The three pinned sites in scripts/test_ci_verify.py and the docs follow. The anchored line PRIVATE_FIXTURE_PROFILE_FLOORS stays at ci_verify.py:802. - docs/ci.md: three sentences still said preflight waits for nothing, although ci.yml now has `needs: changes` and `if: ${{ !cancelled() }}` on preflight (pinned in ci_preflight.py). Reworded on purpose; the old sentences were "Preflight starts at once, and the proof", "`preflight` waits for nothing." and "`changes` and `preflight` wait for nothing and run on every event". Also reworded on purpose: "The two new keys are derived, not measured: each is the registered count in the sampled public pull-request runs" (the kernel key differs from the Release one). The FAST rule gains its one exception (HistoricalProviderPins is skipped without the private overlay and is FAST by decision). - AGENTS.md: "PR Debug, sanitizers and native jobs exclude the measured slow CTest rows" now lists Release and kernel-only too. Note on the first D1 commit message: scripts/release.yml runs no CTest and no ci_verify, so "release keeps every row" there means only that no --exclude-label is present in any release path; the release workflow itself is unchanged by D1. Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
1 parent 1b01244 commit ef8fc5d

4 files changed

Lines changed: 35 additions & 28 deletions

File tree

‎AGENTS.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@ Both C++ unit tests and full corpus verification must pass.
99
The merge ruleset requires `pineforge/verify` and `pineforge/parity` commit
1010
statuses on the PR head's exact tree. The maintainers post them from their own (private)
1111
verification tooling after full verification on their x86-64 build hosts and a
12-
parity verdict; an outside contributor cannot post them. GitHub Actions is advisory: PR Debug, sanitizers and native jobs
12+
parity verdict; an outside contributor cannot post them. GitHub Actions is advisory: PR Release, Debug, kernel-only, sanitizers and native jobs
1313
exclude the measured `slow` CTest rows; push-to-main and manual CI dispatch
1414
run every row, and a documentation-only change runs preflight alone. The commands below run the full sets.
1515

‎docs/ci.md‎

Lines changed: 20 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -733,9 +733,9 @@ of native-live run the registered set excluding
733733
the 29 CTest rows labelled `slow` in `tests/CMakeLists.txt`
734734
([PR tier versus MAIN/RELEASE](#pr-tier-versus-mainrelease-what-is-excluded-where)). The first 27 were
735735
chosen from INT23/INT24 job logs: over 60 seconds in either sanitizer run or
736-
over 30 seconds in either Debug run. Preflight starts at once, and the proof
737-
jobs start alongside it as soon as the few-second `changes` job has decided
738-
whether the change is documentation only. The advisory `build`
736+
over 30 seconds in either Debug run. Preflight and the proof jobs start as soon
737+
as the few-second `changes` job has decided whether the change is
738+
documentation only and which verifier classes a pull request needs. The advisory `build`
739739
aggregate succeeds only if every job succeeds, or was skipped for a
740740
documentation-only change. INT25 re-measured the rows wave G enlarged or added (RATIO-HARDEN's timing
741741
legs, V19-FIX's scaling rows, and the new KERNEL-EDGE, K-ULP4, C-SURFACE-1 and
@@ -802,8 +802,10 @@ parity subset are unchanged.
802802

803803
The verifier's own self-test is split by TestCase class. A class is FAST only
804804
when its measured total in the serial run log is under 5 s and its passing
805-
runtime is measured; every other class is SLOW, and a class in neither list
806-
runs. On a pull request the `verifier-tests` stage of `preflight` runs the FAST
805+
runtime is measured (one exception, by decision: `HistoricalProviderPins` is
806+
skipped without the private overlay, so it has no passing runtime, and is FAST
807+
because it is skipped there); every other class is SLOW, and a class in neither
808+
list runs. On a pull request the `verifier-tests` stage of `preflight` runs the FAST
807809
classes always. It runs the SLOW classes only when `scripts/ci_verifier_paths.py`,
808810
the one place the path list lives, reports that the change touches a path the
809811
verifier reads (`include/pineforge`, `src/c_abi.cpp`, `tests/`, `runner/`, the
@@ -819,11 +821,15 @@ the selection must remove some rows and leave some, and, with the overlay
819821
absent, the rows that ran plus the pinned skips must equal the selection with
820822
no other skip (the `--exclude-label` branch of `enforce_test_floor` in
821823
`scripts/ci_verify.py`). `EXCLUDED_REGISTERED_MIN` holds 876 for `debug`, 867
822-
for `sanitizers` and 963 for `native`, and with D1 895 for `release` and 309
823-
for `kernel`. The two new keys are derived, not measured: each is the
824-
registered count in the sampled public pull-request runs (957 and 371) less the
825-
62 rows by which the Debug key sits below the Debug registry in the same runs
826-
(938 less 876), and the lab job's `ctest -N` confirms them or moves them. The
824+
for `sanitizers` and 963 for `native`, and with D1 895 for `release` and 337
825+
for `kernel`. The two new keys are derived, not measured. `release` is the
826+
registered count in the sampled public pull-request runs (957) less the 62 rows
827+
by which the Debug key sits below the Debug registry in the same runs (938 less
828+
876); that also equals its own registry at the vintage of the other keys. The
829+
same offset would give 309 for `kernel`, 17 rows below the kernel registry at
830+
that vintage (326), so `kernel` takes `KERNEL_MIN_TESTS` (337): 34 below the 371
831+
registered in every sampled pull-request and push run. The lab job's `ctest -N`
832+
confirms both or moves them. The
827833
run floors in the table above (910 and 337 with the overlay present, 603 and
828834
306 absent) are not read on an excluded run; push, dispatch and every other
829835
full run still enforce them.
@@ -913,7 +919,8 @@ and a `changes` job that fails or never answers.
913919
Each proof job waits for `changes` alone, a few seconds once the light ci-spot
914920
class or its hosted fallback takes it, and skips only when `changes` succeeded
915921
and answered documentation-only: `!cancelled() && (needs.changes.result != 'success' ||
916-
needs.changes.outputs.docs_only != 'true')`. `preflight` waits for nothing.
922+
needs.changes.outputs.docs_only != 'true')`. `preflight` waits for `changes`
923+
alone, runs even if `changes` fails (`!cancelled()`), and then runs every verifier class.
917924
The advisory `build` aggregate passes a proof job that succeeded, or one that
918925
was skipped under that same answer. A proof job skipped for any other reason
919926
fails it, and so does any `preflight` result but success. A skipped matrix is
@@ -1083,8 +1090,8 @@ exactly one core-count `--jobs` (counted over the whole job, however its
10831090
commands are chained, blocked or carried, with the flag and its value on one
10841091
line), or any other `--jobs` or `JOBS` value fails preflight, and so does a jobs
10851092
line or `runs-on` the check cannot read; `scripts/test_ci_preflight.py` holds
1086-
the mutations. The same stage pins the documentation-only skip: `changes` and
1087-
`preflight` wait for nothing and run on every event, the classification step
1093+
the mutations. The same stage pins the documentation-only skip: `changes` waits for
1094+
nothing and `preflight` waits for `changes` alone and runs on every event, the classification step
10881095
is pinned verbatim, each proof job needs `changes` alone and carries the one
10891096
skip condition, and the `build` aggregate reads every result and the answer. No
10901097
job of `ci.yml`, `native-live.yml` or `corpus-parity.yml`, nor any of their

‎scripts/ci_verify.py‎

Lines changed: 11 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -645,17 +645,17 @@
645645
# An excluded run must still discover at least this many rows before -LE.
646646
EXCLUDED_REGISTERED_MIN = {
647647
'debug': 876, 'sanitizers': 867, 'native': 963,
648-
# D1: tests-pineforge-engine-ci-pull_request.tsv, run 39221717231 (same
649-
# base for every row). The TSV counts the ctest rows a job SELECTED. The
650-
# Debug and sanitizers PR legs already exclude the 29 `slow` rows, so
651-
# their registered count is selected + 29: Debug 909 + 29 = 938 and
652-
# sanitizers 900 + 29 = 929; offset to their keys 938 - 876 = 62 and
653-
# 929 - 867 = 62. The Release and kernel-only jobs ran their full
654-
# population in that run, so selected = registered: 957 and 371.
655-
# Same registered-count offset: Release 957 - 62 = 895, kernel 371 - 62 = 309.
656-
# Public ctest -N confirmation is still required in the lab; these are
657-
# registration floors, not reduced run floors.
658-
'release': 895, 'kernel': 309,
648+
# D1, run 39221717231 of the PR TSV, REGISTERED counts (Debug/sanitizers
649+
# exclude 29 slow rows, so selected + 29): Debug 938 - 876 = 62 and
650+
# sanitizers 929 - 867 = 62 below their registry; Release and kernel-only
651+
# ran full there: 957 and 371. Release: 957 - 62 = 895. Kernel: the offset
652+
# gives 309, 17 below its registry at that vintage (326), so it takes
653+
# KERNEL_MIN_TESTS 337 (34 below the 371 of every sampled run). The lab's
654+
# ctest -N confirms both; registration floors, not reduced run floors.
655+
# Basis of the offset: the existing keys count rows registered BEFORE -LE,
656+
# so a selected count (909, 900) is never compared with a key (876, 867).
657+
# See docs/ci.md, "PR tier versus MAIN/RELEASE", and COVERAGE-FLOORS.
658+
'release': 895, 'kernel': 337,
659659
}
660660
# test_native_live_websocket exits 77 wherever the linked libcurl has no
661661
# WebSocket support (a system libcurl); main's kernel floor counted that skip

‎scripts/test_ci_verify.py‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1782,7 +1782,7 @@ def test_ctest_label_exclusion_is_forwarded(self):
17821782
def test_pr_exclusion_proves_registered_minus_labelled_equals_ran(self):
17831783
self.assertEqual(ci_verify.EXCLUDED_REGISTERED_MIN,
17841784
{'debug': 876, 'sanitizers': 867, 'native': 963,
1785-
'release': 895, 'kernel': 309})
1785+
'release': 895, 'kernel': 337})
17861786
for profile, registered in ci_verify.EXCLUDED_REGISTERED_MIN.items():
17871787
with self.subTest(profile=profile):
17881788
code, summary, scripted, _ = self.run_profile(
@@ -2817,7 +2817,7 @@ def test_the_table_is_the_floor_and_profiles_outside_it_keep_their_own(self):
28172817
FULL_FLOOR['kernel'])
28182818
self.assertEqual(ci_verify.EXCLUDED_REGISTERED_MIN,
28192819
{'debug': 876, 'sanitizers': 867, 'native': 963,
2820-
'release': 895, 'kernel': 309})
2820+
'release': 895, 'kernel': 337})
28212821

28222822
def test_overlay_flag_moves_the_floor_and_is_the_only_input(self):
28232823
# F9: the flag must MOVE the value. A test that only reads the
@@ -3234,7 +3234,7 @@ def test_release_and_kernel_exclusions_hold_registration_selection_and_pinned_sk
32343234
# D1 floors are independent policy literals, not read back from the
32353235
# implementation being tested. Exercise the real enforcement branch
32363236
# without repeating an engine configure/build or scripted source guard.
3237-
for profile, floor in (('release', 895), ('kernel', 309)):
3237+
for profile, floor in (('release', 895), ('kernel', 337)):
32383238
pinned = self.pinned_probe(1)
32393239
cases = (
32403240
(floor, floor - 5, floor - 5, pinned, 'passed', ''),

0 commit comments

Comments
 (0)