8000 Tags · styk-tv/pgCK · GitHub
[go: up one dir, main page]

Skip to content

Tags

Tags

v0.4.112

Toggle v0.4.112's commit message
Merge remote-tracking branch 'origin/main'

v0.4.111

Toggle v0.4.111's commit message
the upgrade path and the fresh install must mean the same thing

sql/pgck--0.4.109--0.4.111.sql declared ckp.fleet_adoptions STABLE
SECURITY DEFINER; sql/pgck-baseline.sql declared it SECURITY DEFINER.
Same version, two different function contracts depending on whether a
deployment upgraded or installed fresh.

Eight of the nine surface/census verbs in the baseline are VOLATILE
(_adopted_graphs, _composed_shapes, surface_typecheck, surface_unshaped,
surface_declared, surface_grounding, surface_check, fleet_adoptions);
only _engine_counters is STABLE. The migration was the outlier, so it
yields — this changes behaviour nowhere and makes both routes agree.

Root cause is that the migration was hand-written rather than generated
by scripts/gen-upgrade-from-baseline.sh, whose own header records this
defect class twice already (PASS-12 recorded it, PASS-13 measured it,
PASS-PROV a self-inflicted repeat). This is the third. The upgrade path
is exercised by no gate: smoke-s4 and smoke-s34 both CREATE EXTENSION,
never ALTER EXTENSION UPDATE, and the 122 upgrade scripts measured in
the shipped 0.4.109 artifact therefore ship unexecuted.

v0.4.109

Toggle v0.4.109's commit message
Merge branch 'main' of github.com:styk-tv/pgCK

v0.4.108

Toggle v0.4.108's commit message
Merge branch 'main' of github.com:styk-tv/pgCK

v0.4.107

Toggle v0.4.107's commit message
Merge branch 'main' of github.com:styk-tv/pgCK

v0.4.106

Toggle v0.4.106's commit message
Merge branch 'main' of github.com:styk-tv/pgCK

v0.4.105

Toggle v0.4.105's commit message
Merge branch 'main' of github.com:styk-tv/pgCK

v0.4.104

Toggle v0.4.104's commit message
Merge branch 'main' of github.com:styk-tv/pgCK

v0.4.103

Toggle v0.4.103's commit message
feat(instruments): per-kernel storage and the identity triple (v0.4.1…

…03) — C-12 and C-17 flip

C-12: ckp.storage() gains perKernel {kernel: {graphs, asserted, bytes}} —
the data was always attributable (every kernel-owned graph is prefixed
urn:ckp:<kernel>/), the gap was a report, not a mechanism. Shared weight
(core, modules, scratch) lands in a NAMED substrate bucket, never silently
in somebody's column: mis-attributing shared weight manufactures exactly
the busy-vs-bad confusion this closes.

C-17: ckp.identity_triple() — version/build_id/extversion together, the
instrument pgRDF says is the ONLY thing that has ever caught a stale .so
or a pre-tag artifact, and whose absence here let two benches report one
build identifier while running two different binaries. Planes are read
defensively (unreadable = NULL, never a guess) and the verdict comes from
the PURE comparator ckp._identity_agreement — one rule, two callers, so
the probe proves detection against the same code the live verb runs.
Agreement is REFUSED where it cannot be shown ('unmeasurable'); divergence
names both planes and the standing cure: report it, never restart to hide
it. Both instruments surface through surface.check (engineIdentity beside
roster and storage) — a check that is not a verb does not exist.

Controls: s83 (a-f) wired into smoke-s4 — verdict matches hand comparison,
forced mismatch detected naming both planes, missing plane refuses,
fixture graph attributed to its kernel, substrate bucket holds core, both
door-reachable. TDD C-12/C-17 upgraded from existence stubs to behaviour
probes (existence probes can never GREEN — the harness enforces it).

v0.4.102

Toggle v0.4.102's commit message
Merge branch 'main' of github.com:styk-tv/pgCK

0