LTP compliance scan

1,869 cases Β· two rigs Β· per-case verdicts in the full case list β€” reports: docs/test/ltp-report.md, ltp-tracking.md

A full scan of the 1,869-case LTP list (1,838 official binaries cross-compiled with musl, a 101% build rate). Latest official confirmation scan (2026-11-16 scan3, main @ fecf6dd9f, x86_64 + KVM): corrected official PASS 1,138 (60.9%) β€” net +16 over the 2026-11-15 audited 1,122 (all 17 expected gains materialized: 3 in-round scan2 fixes + 11 tail fixes + 3 from ltp-tail2; one real loss, tgkill03); every suspect row was corrected under the isolated-rerun protocol and POISON/WEDGE/NOTRUN stay at zero; the tgkill03 P0 fix (procfs pid-dentry revalidation, 878e3d9f9) is merged to main, so the current expected count is 1,139 (60.9%). Per-case verdicts: the full case list (1,869 rows, the r27 archive, searchable).

PASSTCONFFAILCRASHEXECFAILTIMEOUTWARN
11384342333117124

TCONF is mostly unimplemented feature classes (hugemmap/fanotify/quotactl β€” not failures); FAIL/CRASH/TIMEOUT are the active fronts. The table above is the official corrected distribution of the 2026-11-16 scan3 (zero POISON/WEDGE/NOTRUN rows); the raw r27 distribution lives on the full-case archive page.

Β§1Trend (full scans)

35
524
659
801
945
1005
1028
1010
1059
1093
1122
1138
r1r2r3r9r14r15r20r21r23r2711-1511-16

r1–r10 (riscv64Β·TCG, previous host) β†’ r14–r20 (same test platform, closing at 1028) β†’ r21 onward the x86_64+KVM standing test platform (first round 1010 β†’ r27 audited 1093 β†’ the 2026-11-15 audited round 1122 β†’ the 2026-11-16 official confirmation scan 1138; current expected 1139 with the tgkill03 fix merged). Mid-series dips were mostly host-load noise, recovered through isolated re-runs.

Β§2Methodology

  • Flake isolation: new full-scan failures are isolated-rerun twice before being called regressions β€” host load produces timeout/precision flakes (known: accept02, getrusage04, kill02, genfmod, rename/statfs families).
  • NOT-OURS ledger: the same musl LTP binaries run on the host Linux kernel; identical failures are recorded as test/environment issues rather than kernel gaps (decoded case by case in verdicts-r27.txt).
  • WEDGE rescue: 51 WEDGE/POISON rows from the raw r27 scan were rescued via isolation boots and spliced back; 6 stubborn rows re-ran one-case-per-boot; mmap1 stays WEDGE per the standing adjudication.
  • Every round's full results, deltas and verdicts are archived in that round's work-lane results/ directory (r27 β†’ ltp-r27; the 2026-11-15 audited round β†’ ltp-scan2; the 2026-11-16 confirmation scan β†’ ltp-scan3); TIMEOUT-class cases get top-priority dedicated lanes.

Β§3Test platforms

PlatformLatestNotes
x86_64 + KVM (standing test platform)2026-11-16 scan3 official = 1138/1869; current expected 1139/1869 with the tgkill03 fixthe standing test platform since 2026-11-01; first round r21 = 1010, r27 audited 1093, 11-15 audited 1122
riscv64 Β· TCGr20 = 1028/1869closing round 2026-10-31; r14=945 β†’ r15=1005 (do_exit fix)