Repository navigation
Port the f64 CEL standard library to C++ - #1
Conversation
Register Rust-compatible numeric operations for the C++ parser and runtime, excluding powi and retaining structural sin_cos tuples. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Build Doxygen through cpp-library, link all retained operation contracts to the CEL book, and exercise source and installed consumers. Include repeatable CI, deterministic static analysis, and license notices. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Keep NaN values outside std::clamp's ordering preconditions and validate bounds first. Use Markdown reference input so GitHub and Doxygen resolve the same navigation links. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
sean-parent
left a comment
There was a problem hiding this comment.
Talos local review
Verdict: needs attention. One HIGH-priority operational finding remains open. Six baseline personas completed, followed by the challenge reviewer. This is a COMMENT review, not an approval.
Reviewed feat/f64-port at 7f692612b57df49a6e35b03b7fcc2f3eeec61dfd against main (merge base a507cfe8c78ee652303ecb394f40646a14d783e3).
| Severity | Persona | Total |
|---|---|---|
| HIGH | 1 | 1 |
Change overview
The branch adds the C++20 f64-only CEL standard-library installer, parser/runtime integration, source and installed consumers, documentation, and CI. It retains sin_cos and excludes powi.
Alignment check
The branch and commits align with the f64-port scope. The accepted finding concerns dependency bootstrap reliability, not a numeric result defect. The review did not modify tracked implementation files.
HIGH
| # | Location | Issue | Fix | Source |
|---|---|---|---|---|
| 1 | cmake/CPM.cmake:16 |
A stalled CPM bootstrap transfer has no total or inactivity deadline and can leave initial CMake configuration blocked indefinitely. | Add explicit download TIMEOUT and INACTIVITY_TIMEOUT limits. |
operations |
Coverage and verification
| Files | Reviewers | Candidates -> validated -> accepted | Shown | Tests |
|---|---|---|---|---|
| 27 changed paths; 3 generated templates skipped | 6 personas plus challenge | 1 -> 1 -> 1 | 1 of 1, posted inline | Fresh configure/build and all 4 CTest entries passed |
The personas were security, correctness, standards, architecture, devops, and documentation. Generated-template exclusions: .clang-format, .gitattributes, and docs/index.html. No findings were dropped during validation or duplicate merging. No latent findings, duplicate existing comments, unanchored findings, or inline overflow (accepted 1; posted 1; overflow 0).
Fresh local validation: cmake --preset=test, cmake --build --preset=test --parallel 2, and ctest --preset=test --no-tests=error. The four passing entries cover unit tests, the compiled tutorial example, the independent CPM source consumer, and the installed-package consumer. Existing Linux GCC/Clang, macOS, Windows, analyzer, and documentation CI checks also pass at the reviewed head: https://github.com/stlab/cel-std/actions/runs/43513124864 .
No applicable rule files or GitHub custom instructions existed at the base commit; no active OpenSpec changes were found. Local untracked Scout memory-bank files are outside this review. A passing test run does not exercise a stalled external download.
Workflow adaptations: each persona received the same full, verbatim saved patch as a private read-only artifact rather than an inline duplicate in its prompt. Agents inherited the host's configured models rather than applying the skill's suggested model pins. Scout's section is abbreviated to omit private retrieval counts and local workstation paths.
Persona routing: static roster (DCAP routing off; rerun with --dcap to enable).
Org knowledge: off (rerun with --knowledge).
Scout
Used: risk MEDIUM, evidence HIGH, status COMPLETE (scout 0.9.135).
Scout evidence is incomplete (rows cut by the packet budget); reviewers treated missing evidence as unknown, not as proof.
Scout's listed test surface is tests/consumer/main.cpp, covered by the fresh passing source and installed consumer tests. Cross-file leads were supplied to the evidence-consuming personas and challenge reviewer.
| set(CPM_DOWNLOAD_LOCATION "$ENV{CPM_SOURCE_CACHE}/cpm/CPM_${CPM_DOWNLOAD_VERSION}.cmake") | ||
| else() | ||
| set(CPM_DOWNLOAD_LOCATION "${CMAKE_BINARY_DIR}/cmake/CPM_${CPM_DOWNLOAD_VERSION}.cmake") | ||
| endif() |
There was a problem hiding this comment.
[HIGH] Bound the CPM bootstrap download
When the bootstrap is not already cached and a connection stalls after being established, this file(DOWNLOAD) has neither a total timeout nor an inactivity timeout. CMake configuration can therefore remain blocked indefinitely; the checksum and TLS checks do not impose a deadline. The top-level CMakeLists.txt includes this bootstrap before configuring the library and dependencies, so source consumers are affected too.
Set explicit TIMEOUT and INACTIVITY_TIMEOUT values on this download so a stalled transfer fails within a bounded interval. CMake documents both controls in file(DOWNLOAD).
Verified by the Talos devops and challenge reviewers at 7f692612b57df49a6e35b03b7fcc2f3eeec61dfd.
|
Tracked upstream as cpm-cmake/CPM.cmake#722: cpm-cmake/CPM.cmake#722 . This covers the Talos bootstrap-timeout finding at cmake/CPM.cmake:16. Per the requested disposition, the issue remains tracked upstream rather than fixed in this PR; merging does not imply that the missing download timeouts have been resolved locally. |
Summary
cel::stdlibrary using cpp-library and CPM, with pinnedcel-parser and transitive cel-runtime dependencies.
powi; retain structuralsin_costuples, ordinary/receiver calls, and closure visibility.adapters. Reject invalid clamp bounds, avoid NaN ordering preconditions,
and round ties to even independently of the ambient rounding mode.
signature to the CEL language book. Use Markdown reference input so
GitHub and generated-page navigation both resolve.
approved design, implementation plan, and handoff.
Test plan
Validated locally on macOS at
7f692612b57df49a6e35b03b7fcc2f3eeec61dfd:cmake --preset=test,cmake --build --preset=test --parallel 2,ctest --preset=test --no-tests=error: all 4 tests pass, including theunit executable, tutorial example, CPM source consumer, and installed consumer.
Unit tests contain 12 cases and 248 assertions.
cmake --preset=default,cmake --build --preset=default --parallel 2,and
cmake --install build/default --prefix .../build/pr-install.cmake --preset=clang-tidy,cmake --build --preset=clang-tidy --parallel 2,ctest --preset=clang-tidy --no-tests=error: analyzer and all 4 tests pass.cmake --fresh --preset=docs -G 'Unix Makefiles'andcmake --build --preset=docs: Doxygen warnings treated as errors.The generator override preserves this machine's existing docs build generator;
--freshresets CPM metadata after switching to the preset's source cache.clang-format --dry-run --Werroron every authored C++ source/header.-Wall -Wextra -Wpedantic -Werror;complete consumer suites under AddressSanitizer/UBSan and Ninja Multi-Config.
all 35 published language-book operation anchors verified.
at the same head: Linux GCC/Clang, macOS, Windows, analyzer, and docs.
An independent local branch review identified broken generated reference
navigation; it is fixed and verified. All findings from that review are addressed.
A subsequent Talos local review
completed six personas and the challenge pass at the same head.
It published one HIGH-priority finding at
cmake/CPM.cmake:16:the bootstrap download needs explicit total and inactivity timeouts.
That finding remains open. See the review for its coverage and workflow adaptations.
Generated with GitHub Copilot CLI