Repository navigation
ci(dotnet): split parity cells by layout and retry the SDK download - #1373
Conversation
jev_confidence sums with Python's compensated sum(), which CPython 3.12 introduced; laya allows Python 3.10+, so the last digits match 3.12+ only. Requested on NandhaKishorM#1362.
…eout Each checkpoint now runs in a fused and a split cell (LAYA_TEST_LAYOUTS plus a Layout trait), so every case still runs once while each cell does about half the inference. Per-cell --min-passed floors and timeouts, a runner CPU step, max-parallel 4. Goldens are still regenerated from Python in every cell.
… SDK setup-dotnet failed with HTTP 409 from Microsoft's storage on several jobs on 2026-10-25 (NandhaKishorM#1324 and main), failing them before the build. Retry once after a pause, then use the runner image's .NET 10 SDK, which global.json accepts.
There was a problem hiding this comment.
🟢 Approval recommended
The focused CI and test-selection changes have no identified blocking issues, and reported validation covers all eight jobs and both SDK recovery paths.
No open issues 🎉
What changed in this PR
This PR reduces .NET parity CI duration by dividing checkpoint tests into fused and split layout cells, and improves resilience to SDK download failures.
Changes:
- Add layout selection and test traits, with per-cell pass floors, timeouts, and four-way concurrency.
- Retry SDK installation once, then fall back to a compatible runner-installed SDK.
- Document layout filtering and CPython 3.12+ Jev confidence parity.
| File | Description |
|---|---|
laya-dotnet/tests/Laya.Tests/TestArtifacts.cs |
Define layout traits and environment-based layout selection. |
laya-dotnet/tests/Laya.Tests/ShapeSweepTests.cs |
Tag split-layout shape tests. |
laya-dotnet/tests/Laya.Tests/PredictParityTests.Long.cs |
Filter long-prediction theory rows by layout. |
laya-dotnet/tests/Laya.Tests/PredictParityTests.cs |
Tag split-only prediction tests. |
laya-dotnet/tests/Laya.Tests/PredictParityTests.Batch.cs |
Filter batch-prediction theory rows by layout. |
laya-dotnet/README.md |
Explain parity cells and Python-version numerical parity. |
.github/workflows/dotnet.yml |
Split parity cells, tune execution limits, and add SDK retry and fallback. |
Review effort: Balanced
Resolve the parity timeout conflict with b9fe351 by keeping the per-cell limits, sized from each layout cell's measured time, and carrying that commit's measurements into the comment.
24 PRs. Vietnamese kept off the English checkpoint (#1392), deciding-window attribution preserved through end hooks (#1369), RemoteRouter identifying itself with a configurable User-Agent (#1396), ten more inputs refused where they enter (#1372, #1394, #1397, #1398, #1400, #1401, #1402, #1403, #1404, #1405, #1406), per-language temperatures and per-call budgets in .NET and Rust (#1370, #1371), .NET calibration files (#1384), the Go router lifecycle, SHA-256 pins and routed inference (#1374, #1379, #1395), a C++ foundation (#1399, #1407), and parity-lane sharding that replaces the budget headroom added for 0.4.16 (#1373, #1375). Verified before tagging: all 92 suites pass, 640 tests, up from 612. ruff clean on both lists. laya-ts 1251 tests across 44 files, tsc --noEmit clean. Every generated-artifact gate clean on CPython 3.12 / unicodedata 15.0.0: Java 20 families match laya (9 unverified), UnicodeTables matches CPython 15.0.0, LanguageTables matches laya/lang.py, three Go tables_gen.go match, both laya-cpp .inc files match. Version agrees across pyproject.toml, laya.__version__ and nix/package.nix. README carries 0.4.17 only. Security audit of the merge diff, 362766 added lines over 480 files: zero obfuscation, zero secrets, zero boundary hits, zero binaries, no mode or symlink changes, 20 of 20 actions SHA-pinned including the new cpp.yml, npm install hooks unchanged, and no third-party host in any added line (example.com 90, github.com 2, download.pytorch.org 1, docs.rs 1). The 12 dynamic-exec hits are the C++ test harness invoking its own checker through sys.executable, the Go sample generator reading laya's own example by AST at development time, and one comment in rust.yml. Held: #1364 (unchanged, it moves a cross-port contract), #1381 (a #[doc(hidden)] but pub Rust hook that would ship as public API able to replace the forward pass), #1386 (route_batch switched from self.route to self._route, silently bypassing subclass overrides the docstring promises), and nine stacked port PRs that conflict with each other and need linearizing (#1377, #1378, #1380, #1382, #1383, #1385, #1387, #1388, #1389, #1390). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
Merged. Splitting the parity cells by layout is right, and identifying the SDK download failure as an HTTP 409 from Microsoft's storage is the part that saved me time. I had read that same failure independently and got as far as ruling out a bad pin: I had also raised the parity matrix budget from 60 to 120 minutes just before this landed, after Shipped in 0.4.17. |
Keeps the .NET parity lane inside its time limit after the batch and abstention ports, and stops the Microsoft SDK download outage from failing jobs.
LAYA_TEST_LAYOUTSplus aLayouttrait). Every case still runs once, each cell does about half the inference, and each cell still regenerates its goldens from Python. Per-cell--min-passedfloors and timeouts,max-parallel4, and a runner CPU step.setup-dotnetretries once after a pause, then falls back to the runner's .NET 10 SDK (whichglobal.jsonaccepts). It failed with HTTP 409 from Microsoft's storage on several jobs on 2026-10-25, including test(dotnet): pin Python 0.4.14's English question words and sample its live vocabulary #1324 and main.Tested on our fork: all 8 jobs pass, the longest cell 45 min (typed-decisions 44 and 45, english 35 and 36, multilingual 10 and 14, router 24). On the same runners, untouched main had its typed-decisions cell cancelled at the 60-minute limit and english at 52 min. The retry and fallback were exercised by forcing both download failures on a scratch branch: the retry installed 10.0.403 and the fallback used the runner's 10.0.401, and both jobs passed.