Skip to content

fix(rtc-tcp-client): complete the workflow demo test fixture - #50

Merged
heshaoqiong-tuya merged 3 commits into
masterfrom
feature/improve-ci
Nov 13, 2026
Merged

heshaoqiong-tuya merged 3 commits into
masterfrom
feature/improve-ci

Conversation

@heshaoqiong-tuya

Copy link
Copy Markdown
Collaborator

The demo callbacks dereference dc->action at every EventEnd and the decide path requires a boolean result field; the fixture left action NULL (segfault on GitLab, 0.00 s, no output) and both valid outputs without result, so the capture assertions could never pass either. Set action and add the result field to the streamed chunks, expected document and the valid one-shot output. Test-only, no example changes.

Tests: workflow_variables_demo_test 1/1; full suite 30/30.

The demo callbacks dereference dc->action at every EventEnd and the decide
path requires a boolean result field; the fixture left action NULL (segfault
on GitLab, 0.00 s, no output) and both valid outputs without result, so the
capture assertions could never pass either. Set action and add the result
field to the streamed chunks, expected document and the valid one-shot
output. Test-only, no example changes.

Tests: workflow_variables_demo_test 1/1; full suite 30/30.

Co-Authored-By: Claude Code <noreply@anthropic.com>
@heshaoqiong-tuya
heshaoqiong-tuya force-pushed the feature/improve-ci branch 3 times, most recently from 3b1b71f to e8a7d1f Compare November 9, 2026 12:54
heshaoqiong-tuya and others added 2 commits November 9, 2026 20:58
Build the GitLab side out to mirror .github/workflows/ci.yml: add the
sanitizer and POSIX-examples jobs GitLab was missing, guard the
empty-test-list case the CI-image ctest cannot express (--no-tests=error)
with a portable ctest -N count check, and mark jobs interruptible to
mirror GitHub's cancel-in-progress. Valgrind and pages stay GitLab-only.

Add the CodeScan security scan (tools/codescan_ci.py): the backend clones
the repo itself, so the job only triggers the scan, polls it and applies
the blocking decision -- no build artifacts, riding in the parallel build
wave. Fail-closed severity counting (a platform incident must not read as
a clean bill), credentials stripped before the repo URL leaves the runner,
platform-tunable BLOCK_SEVERITIES, BLOCK_ON_VULN report-only mode.

Review fixes folded in: the parallel wave is safe only under
docker-executor isolation (mock ports 8443/8198/11885 -- keep this
pipeline off shell runners); the examples job asserts every expected demo
target (demos whose prerequisite is missing silently remove themselves
from a green build; audio_chat_demo stays conditional on libopus); the
shared ctest guard and venv/submodule prep live in tools/ci_ctest.sh and
the .venv/.deps hidden jobs with plain YAML anchors.

Verified: YAML parses with the expected six-job structure and every job's
resolved before_script checked; codescan_ci.py against the local API mock
in 5 scenario modes plus a kill-mid-poll run (exit codes and report JSON
as specified); ci_ctest.sh on synthetic ctest trees (clean pass,
empty-list guard, exit-code passthrough); the ctest -N guard checked
against a real 30-test build dir; sanitizer/examples commands
byte-identical to the GitHub jobs already green on ubuntu.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Define the manual contribution workflow for GitLab 17.8.7 CE/Free:
accept GitHub PRs and native GitLab MRs, preserve source SHAs, merge
through one canonical GitLab MR, and fast-forward official branches and
release tags to GitHub. Cover repository protection, isolated validation,
manual review and sync recovery, with matching PR/MR templates. Require
English authoring everywhere except multilingual docs-site content and
the Chinese root README; keep README.md Chinese and README_en.md English;
align AGENTS.md, README entry points and version-tool release guidance.

Then split the docs by audience: CONTRIBUTING.md keeps policy, roles and
the rules every contributor follows (355 -> 161 lines); the runbook keeps
the operational core -- every git command sequence with one-line rules
instead of walkthroughs (213 -> 132 lines, 6 -> 5 sections). An
adversarial pass against the pre-rewrite text confirmed no command line
was lost (14 -> 22 command lines, the additions being the restored import
block) and recovered four rule-level items the first compression dropped.
Drop the never-defined "Mode A" label (5 uses, 0 definitions, no Mode B
exists) and update the references that pointed at renamed headings.

Record the workflow docs and the pipeline additions under [Unreleased]
in CHANGELOG.md, linked to the canonical MR !5 (AGENTS.md reference
rules). The pre-existing workflow_variables_demo "(PR pending)" entry
predates this branch (it is on master) and is not this change's debt.

Validation: 27/27 local Markdown links resolved; 16/16 shell examples
parsed without execution; version-tool Bash syntax, language policy and
diff whitespace checks passed; command-inventory diff across the runbook
rewrite, balanced fences, resolving cross-references, every repo
reference to the runbook truthful after the restructure, English-only
compliance. SDK tests run: 0 (no SDK changes).

Co-Authored-By: Codex <noreply@openai.com>
Co-Authored-By: Claude Code <noreply@anthropic.com>
@heshaoqiong-tuya
heshaoqiong-tuya merged commit 3d6e680 into master Nov 13, 2026
8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant