Skip to content

Commit 9151a11

Browse files
committed
the published reading is builds, and mine was runs
CI measured mimalloc's Windows cell as `builds`: the backend crash is gone and the member compiles and links, and its `alloc` test then exits 1 under the Wine that CI installs while passing under the one on this machine. The recipe comment claimed `runs` on the strength of the local run. The build repair stands — the cell was `fails: error: build failed` — but a repaired build is not a passing test, and both the recipe and docs now say which one this is. The sixty-cell table is CI's, not mine: 56 runs, 1 builds, 3 fails, all three declared.
1 parent d1e6f17 commit 9151a11

2 files changed

Lines changed: 29 additions & 0 deletions

File tree

docs/openkal-compat.md

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -167,6 +167,25 @@ lower. The comparison becomes a required check for pull requests once the
167167
repository variable `OPENKAL_RATCHET` is `on`; it is enabled after the weekly
168168
measurement has been stable for two consecutive weeks.
169169

170+
### The 2026-09-21 measurement, on mcpp 2026.9.21.3
171+
172+
Thirty members, two targets, sixty cells, measured by this repository's own CI:
173+
174+
| | before | after |
175+
| --- | --- | --- |
176+
| `runs` | 50 | **56** |
177+
| `builds` | 0 | 1 |
178+
| `fails` | 10 | **3, all declared** |
179+
180+
`compat.py check` against the published baseline: no published label
181+
regressed, and no declaration was contradicted.
182+
183+
The one `builds` is `mimalloc` on `x86_64-windows-gnu`. Its build was repaired
184+
in this round — `MI_USE_BUILTIN_THREAD_POINTER=0`, see the recipe — and its
185+
`alloc` test then exits 1 under the Wine CI installs while passing under
186+
another. A repaired build is not a passing test, and the label says which one
187+
this is.
188+
170189
### Why `aarch64-macos` is not a pinned target yet
171190

172191
`pins.toml` names linux and windows. macOS is measurable — `mcpp test --no-run`

pkgs/c/compat.mimalloc.lua

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -135,6 +135,16 @@ package = {
135135
-- Windows branch, and on a Win32 target the setting is not read at
136136
-- all --- `_WIN32` selects `NtCurrentTeb()` three arms earlier ---
137137
-- so it changes only the target that has no `_WIN32`.
138+
--
139+
-- WHAT THIS FIXES AND WHAT IT DOES NOT. The cell goes from
140+
-- `fails: error: build failed` to `builds`: the backend crash is
141+
-- gone and the member compiles and links for this target. Its
142+
-- `alloc` test then passes under one Wine and exits 1 under the
143+
-- one this repository's CI installs, so the published reading is
144+
-- `builds` rather than `runs`. That remainder is a separate
145+
-- question about the TLS path the override falls back to, and it
146+
-- is recorded here rather than folded into this one so the next
147+
-- reader does not take a green build for a green test.
138148
cflags = { "-DMI_USE_BUILTIN_THREAD_POINTER=0" },
139149
},
140150
},

0 commit comments

Comments
 (0)