debug-tools: Build and package amd-dbgapi test suite by lumachad · Pull Request #8943 · ROCm/TheRock

lumachad · 2026-10-12T16:29:08Z

Summary

rocdbgapi now builds a unit test suite upstream, gated by ENABLE_TESTS and ENABLE_UNIT_TESTS, and installs the test binaries under a dedicated test install component. THEROCK_BUILD_TESTING is wired through to ENABLE_TESTS so the suite is built whenever testing is enabled, and therock-googletest is added as a build dependency only in that case.

The test binaries are folded into the existing amd-dbgapi artifact as a new test component rather than a separate artifact, matching the pattern already used by rocgdb. They are excluded from the dev component so they only ship once. build_tools/packaging/linux/package.json was updated so the new component is covered by the amdrocm-debugger-test package.

The committed consumer graph was updated with the new amd-dbgapi to therock-googletest build dependency edge.

This PR only builds and packages the tests. Running them in CI is left for a follow-up.

ISSUE ID: AIROCGDB-634

🤖 Generated with Claude Code

therock-pr-bot · 2026-10-12T16:29:51Z

✅ All Checks Passed — Ready for Review

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	—
🔎 pre-commit	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—
🤖 therock-pr-bot	✅ Pass	—

🎉 All checks passed! This PR is ready for review.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

therock-pr-bot · 2026-10-12T16:29:53Z

🎉 All checks passed! This PR is ready for review.

amd-bfilipov

Looks good, thanks.

marbre

------------------------------ Captured log call -------------------------------
ERROR    test_artifact_structure:test_artifact_structure.py:452 MISSING package coverage: amd-dbgapi [test] (amd-dbgapi_test_generic.tar.zst)
=========================== short test summary info ============================
FAILED tests/test_artifact_structure.py::TestArtifactStructure::test_artifacts_covered_by_packages - Failed: Found 1 artifact component(s) not covered by package.json:
  amd-dbgapi [test] (amd-dbgapi_test_generic.tar.zst)
========================= 1 failed, 2 passed in 10.46s =========================

Can you please address this @lumachad?

lumachad · 2026-10-20T10:21:16Z

------------------------------ Captured log call -------------------------------
ERROR    test_artifact_structure:test_artifact_structure.py:452 MISSING package coverage: amd-dbgapi [test] (amd-dbgapi_test_generic.tar.zst)
=========================== short test summary info ============================
FAILED tests/test_artifact_structure.py::TestArtifactStructure::test_artifacts_covered_by_packages - Failed: Found 1 artifact component(s) not covered by package.json:
  amd-dbgapi [test] (amd-dbgapi_test_generic.tar.zst)
========================= 1 failed, 2 passed in 10.46s =========================

Can you please address this @lumachad?

Will do. Thanks for spotting that.

motokultivator

Looks good to me.

Split rocr-debug-agent CI configuration into CPU and GPU test jobs, following the pattern established for amd-dbgapi in #8943. - rocr-debug-agent-cpu: CPU runner, placeholder test script - rocr-debug-agent-gpu: GPU runner, placeholder test script Both jobs use echo placeholders that always pass until the test wrapper script is updated with --cpu-tests and --gpu-tests flags. Added rocr-debug-agent to TEST_LABEL_GROUPS so test:rocr-debug-agent label triggers both variants. JIRA ID: AIROCGDB-635

nunnikri

LGTM

peymanr · 2026-10-30T17:10:07Z

PR Pundit review

(Testing PR Pundit Code Review agent. Not all comments may be relevant. This is a one-time automated pass: please address what is useful, no need to request a re-review. A quick reply per point (useful or not, and why) helps us calibrate the agent. Reach out to @peymanr for suggestions and feedback.)

#		Finding	Where
1	🟠	Echo-only amd-dbgapi placeholder jobs scheduled by default, wasting GPU runners: this PR added ungated matrix entries	`build_tools/github_actions/configure_multi_arch_ci.py:188`
2	🟠	amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only `_rocgdb_common` jobs	`build_tools/github_actions/fetch_test_configurations.py:555-570`
3	🟠	Unit test pins placeholder `echo` script, forcing edits when real runner lands: this PR asserts on placeholder text	`build_tools/github_actions/tests/fetch_test_configurations_test.py:1285`

1. 🟠 Should fix: Echo-only amd-dbgapi placeholder jobs scheduled by default, wasting GPU runners: this PR added ungated matrix entries

build_tools/github_actions/configure_multi_arch_ci.py:188

amd-dbgapi-cpu/amd-dbgapi-gpu only run echo and always pass, yet nothing gates them (no test_types, include_family, or label), so every default CI run (PROJECTS_TO_TEST='*') schedules them on every Linux and Windows family. amd-dbgapi-gpu has no linux_cpu_runner, so it occupies a scarce GPU runner (e.g. linux-gfx942-1gpu-ccs-csp-ossci-rocm, windows-gfx1151-gpu-rocm) just to download --debug-tools --tests artifacts and print a message, adding a green check with zero coverage. It also drops rocgdb-gpu's gfx125X-dcgpu exclude_family that a real runner will need. Gate them to label-only (or omit amd-dbgapi-gpu) until a real runner script exists.

Who is affected: Every Linux and Windows CI test run with debug-tools artifacts schedules two no-op jobs per GPU family.

Checked against the code by an independent pass.

Related past reviews: #2231 on debug-tools/CMakeLists.txt, #2859 on build_tools/github_actions/test_executable_scripts/test_rocr-debug-agent.py, #4260 on build_tools/github_actions/test_executable_scripts/test_rocr-debug-agent.py

2. 🟠 Should fix: amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only `_rocgdb_common` jobs

build_tools/github_actions/fetch_test_configurations.py:555-570

amd-dbgapi-cpu/-gpu reuse the Linux-oriented _rocgdb_common (rocgdb Ubuntu image, --cap-add=SYS_PTRACE, --debug-tools --tests) and override platform to ['linux','windows'] without a comment. On Windows the container settings are discarded, as they are for every job. --debug-tools also requests rocgdb/rocr-debug-agent* artifacts that are not built on Windows (disable_platforms=['windows']); missing patterns are skipped silently. amd-dbgapi itself does build on Windows, so enabling Windows is plausible. A short comment explaining the Windows opt-in, or narrowing the fetch args, would make the intent clear.

Who is affected: Windows CI runs that select the debug-tools stage or test:amd-dbgapi; they schedule placeholder jobs that just echo and pass.

Checked against the code by an independent pass.

Related past reviews: #2540 on build_tools/install_rocm_from_artifacts.py, #2797 on build_tools/fetch_sources.py, #2051 on debug-tools/CMakeLists.txt, #3379 on build_tools/github_actions/fetch_test_configurations.py, #6436 on build_tools/github_actions/fetch_test_configurations.py, #8689 on build_tools/github_actions/configure_multi_arch_ci.py, #8710 on build_tools/github_actions/configure_multi_arch_ci.py, #8768 on build_tools/github_actions/configure_multi_arch_ci.py

3. 🟠 Should fix: Unit test pins placeholder `echo` script, forcing edits when real runner lands: this PR asserts on placeholder text

build_tools/github_actions/tests/fetch_test_configurations_test.py:1285

test_amd_dbgapi_cpu_job_runs_on_cpu_runner_with_placeholder_script asserts assertIn('echo', job['test_script']), which pins the temporary placeholder script. Anyone replacing the placeholder with a real runner must also delete this assertion. Its only other check is linux_cpu_runner=True, a static flag. That flag is already exercised for CPU-only jobs by test_cpu_only_components_run_without_gpu_runner. Consider dropping the script-text assertion, or folding the CPU-runner check into that existing test.

Who is affected: Maintainers who later wire up a real amd-dbgapi test runner; no CI or user behaviour is affected.

Checked against the code by an independent pass.

Related past reviews: #8319 on build_tools/github_actions/tests/configure_multi_arch_ci_test.py, #4616 on build_tools/github_actions/tests/configure_multi_arch_ci_test.py, #8169 on build_tools/github_actions/tests/configure_multi_arch_ci_test.py

_{Shared from a pilot of PR Pundit, an agentic code reviewer that builds and tests the change and reads the project's past reviews. Reviewed at 79a2f53e; findings checked and posted by @peymanr; points already raised on this PR are left out.}

lumachad · 2026-11-02T12:44:02Z

PR Pundit review

(Testing PR Pundit Code Review agent. Not all comments may be relevant. This is a one-time automated pass: please address what is useful, no need to request a re-review. A quick reply per point (useful or not, and why) helps us calibrate the agent. Reach out to @peymanr for suggestions and feedback.)

Finding Where

1 🟠 Echo-only amd-dbgapi placeholder jobs scheduled by default, wasting GPU runners: this PR added ungated matrix entries build_tools/github_actions/configure_multi_arch_ci.py:188
2 🟠 amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only _rocgdb_common jobs build_tools/github_actions/fetch_test_configurations.py:555-570
3 🟠 Unit test pins placeholder echo script, forcing edits when real runner lands: this PR asserts on placeholder text build_tools/github_actions/tests/fetch_test_configurations_test.py:1285

1. 🟠 Should fix: Echo-only amd-dbgapi placeholder jobs scheduled by default, wasting GPU runners: this PR added ungated matrix entries

build_tools/github_actions/configure_multi_arch_ci.py:188

amd-dbgapi-cpu/amd-dbgapi-gpu only run echo and always pass, yet nothing gates them (no test_types, include_family, or label), so every default CI run (PROJECTS_TO_TEST='*') schedules them on every Linux and Windows family. amd-dbgapi-gpu has no linux_cpu_runner, so it occupies a scarce GPU runner (e.g. linux-gfx942-1gpu-ccs-csp-ossci-rocm, windows-gfx1151-gpu-rocm) just to download --debug-tools --tests artifacts and print a message, adding a green check with zero coverage. It also drops rocgdb-gpu's gfx125X-dcgpu exclude_family that a real runner will need. Gate them to label-only (or omit amd-dbgapi-gpu) until a real runner script exists.

Who is affected: Every Linux and Windows CI test run with debug-tools artifacts schedules two no-op jobs per GPU family.

Checked against the code by an independent pass.

Related past reviews: #2231 on debug-tools/CMakeLists.txt, #2859 on build_tools/github_actions/test_executable_scripts/test_rocr-debug-agent.py, #4260 on build_tools/github_actions/test_executable_scripts/test_rocr-debug-agent.py

2. 🟠 Should fix: amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only _rocgdb_common jobs

build_tools/github_actions/fetch_test_configurations.py:555-570

amd-dbgapi-cpu/-gpu reuse the Linux-oriented _rocgdb_common (rocgdb Ubuntu image, --cap-add=SYS_PTRACE, --debug-tools --tests) and override platform to ['linux','windows'] without a comment. On Windows the container settings are discarded, as they are for every job. --debug-tools also requests rocgdb/rocr-debug-agent* artifacts that are not built on Windows (disable_platforms=['windows']); missing patterns are skipped silently. amd-dbgapi itself does build on Windows, so enabling Windows is plausible. A short comment explaining the Windows opt-in, or narrowing the fetch args, would make the intent clear.

Who is affected: Windows CI runs that select the debug-tools stage or test:amd-dbgapi; they schedule placeholder jobs that just echo and pass.

Checked against the code by an independent pass.

Related past reviews: #2540 on build_tools/install_rocm_from_artifacts.py, #2797 on build_tools/fetch_sources.py, #2051 on debug-tools/CMakeLists.txt, #3379 on build_tools/github_actions/fetch_test_configurations.py, #6436 on build_tools/github_actions/fetch_test_configurations.py, #8689 on build_tools/github_actions/configure_multi_arch_ci.py, #8710 on build_tools/github_actions/configure_multi_arch_ci.py, #8768 on build_tools/github_actions/configure_multi_arch_ci.py

3. 🟠 Should fix: Unit test pins placeholder echo script, forcing edits when real runner lands: this PR asserts on placeholder text

build_tools/github_actions/tests/fetch_test_configurations_test.py:1285

test_amd_dbgapi_cpu_job_runs_on_cpu_runner_with_placeholder_script asserts assertIn('echo', job['test_script']), which pins the temporary placeholder script. Anyone replacing the placeholder with a real runner must also delete this assertion. Its only other check is linux_cpu_runner=True, a static flag. That flag is already exercised for CPU-only jobs by test_cpu_only_components_run_without_gpu_runner. Consider dropping the script-text assertion, or folding the CPU-runner check into that existing test.

Who is affected: Maintainers who later wire up a real amd-dbgapi test runner; no CI or user behaviour is affected.

Checked against the code by an independent pass.

Related past reviews: #8319 on build_tools/github_actions/tests/configure_multi_arch_ci_test.py, #4616 on build_tools/github_actions/tests/configure_multi_arch_ci_test.py, #8169 on build_tools/github_actions/tests/configure_multi_arch_ci_test.py

Shared from a pilot of PR Pundit, an agentic code reviewer that builds and tests the change and reads the project's past reviews. Reviewed at 79a2f53e; findings checked and posted by @peymanr; points already raised on this PR are left out.

Fair enough. I'm leaving the test enablement in CI for a follow-up.

lumachad · 2026-11-04T09:49:53Z

I'd like to merge this reduced scope PR later today. Do let me know if you have objections.

arvindcheru · 2026-11-04T16:48:00Z

      amd-llvm
    CMAKE_ARGS
      -DTHE_ROCK_BUILD=ON
+      "-DENABLE_TESTS=${THEROCK_BUILD_TESTING}"


Thanks for the changes @lumachad, can ENABLE_TESTS stay off until an upstream unit test actually exercises amd-dbgapi? For my understanding: the new binary dbgapi_smoke_test only links GTest::gtest_main and GTest::gmock. It does not call amd-dbgapi. Does the follow-up add that unit test, or does it only run this smoke binary?

can ENABLE_TESTS stay off until an upstream unit test actually exercises amd-dbgapi?

I'd rather leave the ENABLE_TESTS gated on THEROCK_BUILD_TESTING, as that means we always go through the infra to build these and install the files. That way we'll always test this path when we bump rocm-systems, even though we don't have a complete set of unit tests or invoke them.

For my understanding: the new binary dbgapi_smoke_test only links GTest::gtest_main and GTest::gmock. It does not call amd-dbgapi. Does the follow-up add that unit test, or does it only run this smoke binary?

The follow-up will start running these CPU-based smoke tests first, then the set will be expanded to do real CPU-based tests. Then we will eventually have GPU-based ones.

Enablement of the unit tests is taking place in parallel, and will be done in incremental steps.

Does that answer your questions?

Thank you . This explains why ENABLE_TESTS stays tied to THEROCK_BUILD_TESTING.
And reconfirms The smoke binary, the missing amd-dbgapi unit test, and the later GPU tests are expected in follow-up

@lumachad at this time dbgapi_smoke_test is included in packaging, but not getting tested with CI, right?
Would you have a link to dbgapi_smoke_test test results to make sure it runs fine?

@dileepr1 That's correct. I did the validation locally on Linux. The Windows runners had some hiccups that prevented running it. I think I can put up a second temporary commit to at least confirm the smoke test runs fine on both Windows and Linux. Would that be acceptable?

Once we see it working, I can drop the temporary patch and go with just building it but not running it in CI.

That would be great, thanks!

Pushed a temporary commit now.

@dileepr1 @arvindcheru Here's one of the Windows smoke test jobs running: https://github.com/ROCm/TheRock/actions/runs/42214365465/job/127909265685?pr=8943

And the Linux one: https://github.com/ROCm/TheRock/actions/runs/42214365465/job/127910206450?pr=8943

rocdbgapi now builds a unit test suite through ENABLE_TESTS and ENABLE_UNIT_TESTS, installing the test binaries under a dedicated install component. Wire THEROCK_BUILD_TESTING through to ENABLE_TESTS so the suite is built whenever testing is enabled, add therock-googletest as a build dependency when that's the case, and fold the test binaries into the amd-dbgapi artifact as a new test component so they get staged and distributed like everything else. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

Temporary patch to try the dbgapi-cpu test job in CI. Adds a test matrix entry on the rocgdb CPU container and a script that runs each GoogleTest binary installed under tests/rocdbgapi. Not meant to merge with the test build PR. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

github-project-automation Bot added this to TheRock Triage Oct 12, 2026

lumachad marked this pull request as ready for review October 12, 2026 16:29

lumachad requested review from a team as code owners October 12, 2026 16:29

therock-pr-bot Bot added the Not ready to Review PR has unresolved policy failures — reviews blocked label Oct 12, 2026

lumachad requested a review from marbre October 12, 2026 16:30

therock-pr-bot Bot removed the Not ready to Review PR has unresolved policy failures — reviews blocked label Oct 12, 2026

lumachad mentioned this pull request Oct 12, 2026

Unit & feature tests infrastructure ROCm/rocm-systems#12199

Merged

5 tasks

amd-bfilipov approved these changes Oct 15, 2026

View reviewed changes

marbre reviewed Oct 19, 2026

View reviewed changes

lumachad marked this pull request as draft October 20, 2026 10:21

lumachad force-pushed the users/lumachad/dbgapi-tests branch from 40e8201 to 7770bec Compare October 20, 2026 12:51

lumachad changed the title ~~debug-tools: Build and package amd-dbgapi test suite~~ debug-tools: Build amd-dbgapi test suite and wire up placeholder CI jobs Oct 20, 2026

lumachad force-pushed the users/lumachad/dbgapi-tests branch from 7770bec to 5d19d13 Compare October 23, 2026 08:46

lumachad mentioned this pull request Oct 23, 2026

debug-tools: Split rocr-debug-agent tests into CPU and GPU jobs #9357

Open

lumachad force-pushed the users/lumachad/dbgapi-tests branch from 5d19d13 to 8d84836 Compare October 26, 2026 09:49

motokultivator approved these changes Oct 26, 2026

View reviewed changes

lumachad force-pushed the users/lumachad/dbgapi-tests branch 2 times, most recently from 0bbf069 to 0d65574 Compare October 29, 2026 08:54

lumachad marked this pull request as ready for review October 29, 2026 08:54

lumachad requested review from arvindcheru, nunnikri and raramakr as code owners October 29, 2026 08:54

nunnikri approved these changes Oct 29, 2026

View reviewed changes

lumachad force-pushed the users/lumachad/dbgapi-tests branch from 0d65574 to 79a2f53 Compare October 30, 2026 10:27

marbre reviewed Oct 30, 2026

View reviewed changes

Comment thread build_tools/github_actions/fetch_test_configurations.py Resolved

lumachad force-pushed the users/lumachad/dbgapi-tests branch from 79a2f53 to e790c61 Compare October 30, 2026 15:48

lumachad requested review from ScottTodd and jayhawk-commits as code owners October 30, 2026 15:48

lumachad force-pushed the users/lumachad/dbgapi-tests branch from e790c61 to 48b1085 Compare October 30, 2026 15:49

lumachad added the ci:skip Skip all CI builds/tests for this PR label Oct 30, 2026

lumachad force-pushed the users/lumachad/dbgapi-tests branch from 48b1085 to c52ed31 Compare November 2, 2026 12:42

lumachad changed the title ~~debug-tools: Build amd-dbgapi test suite and wire up placeholder CI jobs~~ debug-tools: Build and package amd-dbgapi test suite Nov 2, 2026

ScottTodd removed their request for review November 2, 2026 19:28

lumachad removed the ci:skip Skip all CI builds/tests for this PR label Nov 3, 2026

lumachad force-pushed the users/lumachad/dbgapi-tests branch from c52ed31 to b9209a5 Compare November 3, 2026 14:11

arvindcheru reviewed Nov 4, 2026

View reviewed changes

lumachad added ci:skip Skip all CI builds/tests for this PR and removed ci:skip Skip all CI builds/tests for this PR labels Nov 6, 2026

lumachad and others added 2 commits November 6, 2026 12:00

lumachad force-pushed the users/lumachad/dbgapi-tests branch from c505c71 to b9e5b8d Compare November 6, 2026 12:00

Conversation

lumachad commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Uh oh!

therock-pr-bot Bot commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ All Checks Passed — Ready for Review

Uh oh!

therock-pr-bot Bot commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

amd-bfilipov left a comment

Choose a reason for hiding this comment

Uh oh!

marbre left a comment

Choose a reason for hiding this comment

Uh oh!

lumachad commented Oct 20, 2026

Uh oh!

motokultivator left a comment

Choose a reason for hiding this comment

Uh oh!

nunnikri left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

peymanr commented Oct 30, 2026

PR Pundit review

1. 🟠 Should fix: Echo-only amd-dbgapi placeholder jobs scheduled by default, wasting GPU runners: this PR added ungated matrix entries

2. 🟠 Should fix: amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only _rocgdb_common jobs

3. 🟠 Should fix: Unit test pins placeholder echo script, forcing edits when real runner lands: this PR asserts on placeholder text

Uh oh!

lumachad commented Nov 2, 2026

PR Pundit review

Finding Where

1. 🟠 Should fix: Echo-only amd-dbgapi placeholder jobs scheduled by default, wasting GPU runners: this PR added ungated matrix entries

2. 🟠 Should fix: amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only _rocgdb_common jobs

3. 🟠 Should fix: Unit test pins placeholder echo script, forcing edits when real runner lands: this PR asserts on placeholder text

Uh oh!

lumachad commented Nov 4, 2026

Uh oh!

arvindcheru Nov 4, 2026

Choose a reason for hiding this comment

Uh oh!

lumachad Nov 4, 2026

Choose a reason for hiding this comment

Uh oh!

arvindcheru Nov 4, 2026

Choose a reason for hiding this comment

Uh oh!

dileepr1 Nov 4, 2026

Choose a reason for hiding this comment

Uh oh!

lumachad Nov 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dileepr1 Nov 5, 2026

Choose a reason for hiding this comment

Uh oh!

lumachad Nov 6, 2026

Choose a reason for hiding this comment

Uh oh!

lumachad Nov 6, 2026

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

lumachad commented Oct 12, 2026 •

edited

Loading

therock-pr-bot Bot commented Oct 12, 2026 •

edited

Loading

therock-pr-bot Bot commented Oct 12, 2026 •

edited

Loading

2. 🟠 Should fix: amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only `_rocgdb_common` jobs

3. 🟠 Should fix: Unit test pins placeholder `echo` script, forcing edits when real runner lands: this PR asserts on placeholder text

2. 🟠 Should fix: amd-dbgapi Windows jobs fetch nonexistent artifact patterns: this PR added Windows to Linux-only `_rocgdb_common` jobs

3. 🟠 Should fix: Unit test pins placeholder `echo` script, forcing edits when real runner lands: this PR asserts on placeholder text

lumachad Nov 5, 2026 •

edited

Loading