From f2e4457f97461056f7a7f3a3249d97d109ee788a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Deniz=20=C4=B0rgin?= <denizirgin@gmail.com>
Date: Sat, 17 Oct 2026 17:03:09 +0300
Subject: [PATCH] ci: a hung test module fails inside its pass under a session
 timeout (#447)

Each test pass runs with a 25-minute --timeout, about four times the slowest
measured module, and a failed main pass also writes its test listing, so the
rows still in flight are among those listed but not recorded.
---
 .github/workflows/ci.yml          | 14 +++++++++++---
 docs/engineering/quality-gates.md |  7 +++++++
 2 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 3c3ee033..454cef24 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -301,14 +301,22 @@ jobs:
       # reason. They run after a failed first pass (not after a failed build), so the report sees
       # every pass, and they leave the HTML report upload to the first pass, whose artifact names
       # (assembly, OS, framework) they would repeat.
+      #
+      # Every pass ends a hung module inside its step under a session --timeout; the listing below,
+      # written only when the main pass fails, narrows the rows in flight (quality-gates.md).
       - name: Test
-        run: dotnet test --configuration Release --no-build --max-parallel-test-modules ${{ matrix.test-modules }} -- --report-trx --results-directory "${{ github.workspace }}/test-results/main" --treenode-filter "/*/*/*/*[(Category!=ServiceElection)&(Category!=NativeAot)]" --minimum-expected-tests 1
+        id: test
+        run: dotnet test --configuration Release --no-build --max-parallel-test-modules ${{ matrix.test-modules }} -- --report-trx --results-directory "${{ github.workspace }}/test-results/main" --treenode-filter "/*/*/*/*[(Category!=ServiceElection)&(Category!=NativeAot)]" --minimum-expected-tests 1 --timeout 25m
         env:
           TUNIT_ARTIFACT_RETENTION_DAYS: 7
 
+      - name: List tests (main, after a failed pass)
+        if: ${{ !cancelled() && steps.test.outcome == 'failure' }}
+        run: dotnet test --configuration Release --no-build --list-tests -- --treenode-filter "/*/*/*/*[(Category!=ServiceElection)&(Category!=NativeAot)]" > "${{ github.workspace }}/test-results/main/listed-tests.txt"
+
       - name: Test (service elections)
         if: ${{ !cancelled() && steps.build.outcome == 'success' }}
-        run: dotnet test --project "${{ github.workspace }}/tests/OpenCode.Sdk.Tests/OpenCode.Sdk.Tests.csproj" --configuration Release --no-build --max-parallel-test-modules 1 -- --report-trx --results-directory "${{ github.workspace }}/test-results/service-elections" --treenode-filter "/*/*/*/*[Category=ServiceElection]" --minimum-expected-tests 1
+        run: dotnet test --project "${{ github.workspace }}/tests/OpenCode.Sdk.Tests/OpenCode.Sdk.Tests.csproj" --configuration Release --no-build --max-parallel-test-modules 1 -- --report-trx --results-directory "${{ github.workspace }}/test-results/service-elections" --treenode-filter "/*/*/*/*[Category=ServiceElection]" --minimum-expected-tests 1 --timeout 25m
         env:
           TUNIT_ARTIFACT_RETENTION_DAYS: 7
           TUNIT_DISABLE_ARTIFACT_UPLOAD: true
@@ -318,7 +326,7 @@ jobs:
       # host's native toolchain and runs the host's own runtime identifier.
       - name: Test (native AOT)
         if: ${{ !cancelled() && steps.build.outcome == 'success' }}
-        run: dotnet test --project "${{ github.workspace }}/tests/OpenCode.Sdk.Tools.Tests/OpenCode.Sdk.Tools.Tests.csproj" --configuration Release --no-build --max-parallel-test-modules 1 -- --report-trx --results-directory "${{ github.workspace }}/test-results/native-aot" --treenode-filter "/*/*/*/*[Category=NativeAot]" --minimum-expected-tests 1
+        run: dotnet test --project "${{ github.workspace }}/tests/OpenCode.Sdk.Tools.Tests/OpenCode.Sdk.Tools.Tests.csproj" --configuration Release --no-build --max-parallel-test-modules 1 -- --report-trx --results-directory "${{ github.workspace }}/test-results/native-aot" --treenode-filter "/*/*/*/*[Category=NativeAot]" --minimum-expected-tests 1 --timeout 25m
         env:
           TUNIT_ARTIFACT_RETENTION_DAYS: 7
           TUNIT_DISABLE_ARTIFACT_UPLOAD: true
diff --git a/docs/engineering/quality-gates.md b/docs/engineering/quality-gates.md
index ec100c9f..335290bd 100644
--- a/docs/engineering/quality-gates.md
+++ b/docs/engineering/quality-gates.md
@@ -230,6 +230,13 @@ no unexpected or duplicate modules. The roster is deliberately explicit rather t
 cannot shrink together with the coverage it guards — so adding or removing a test project or a test
 target framework updates it in the same change.
 
+Every CI test pass runs under a 25-minute session `--timeout`, about four times the slowest
+measured module (`OpenCode.Sdk.Tests` on net472 on Windows, under six minutes), so a hung module
+fails inside its step instead of at the job timeout. An expired session exits with code 7 and keeps
+the TRX of the rows that finished, but it does not name the rows still in flight. When the main
+pass fails, the job also writes that pass's test listing to `test-results/main/listed-tests.txt`,
+and the rows in flight are among those listed but not recorded.
+
 ## Clone gate
 
 The detector ignores using directives; a baseline entry that matches no clone fails the gate like a new clone does.