Repository navigation
Windows: name CMake intermediate dirs by short hashes (CMAKE_INTERMEDIATE_DIR_STRATEGY=SHORT) - #178
Merged
Tobias-Fischer merged 1 commit intoOct 22, 2026
Conversation
Long package names push MSVC object paths past 260 characters, e.g. rosbag2_performance_benchmarking_msgs' generated type support failed with C1083: Cannot open compiler generated file: '' in RoboStack's jazzy and lyrical win-64 builds. CMAKE_INTERMEDIATE_DIR_STRATEGY=SHORT (CMake >= 4.2, Ninja and Visual Studio generators) names the intermediate directories and object files by short hashes. It has to be a CMake argument: rattler-build doesn't pass the environment variable through to the build script. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
traversaro
approved these changes
Oct 22, 2026
Tobias-Fischer
added a commit
to RoboStack/robostack
that referenced
this pull request
Oct 23, 2026
vinca's Windows build scripts pass -DCMAKE_INTERMEDIATE_DIR_STRATEGY=SHORT for every package since RoboStack/vinca#178 (pinned with #139, which has to be merged before this PR). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tobias-Fischer
added a commit
to RoboStack/robostack
that referenced
this pull request
Oct 23, 2026
* lyrical: on the shared conda-forge pinning, Qt6 mapping shared (from #82) Drops the distribution's own conda-forge-pinning version and migrations: it uses shared/pinning/conda_forge.yaml (2026.10.09.13.11.31), with the package fixes the new pins needed. Other distributions keep their own pinning version until their PR; their conda_build_config.yaml only gets the comment of the new shared openexr override (the value, 3.4, is what they pin already). shared/robostack.yaml maps the Qt5 rosdep keys to Qt6 (qt6-main, pyqt6), so lyrical's per-package Qt workarounds go. humble and jazzy keep Qt5 through overrides that their own sync PRs add: merge those first. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * lyrical: rosdistro sync 2026-10-05 `rs lyrical update-snapshot`, with the fixes from the full-rebuild runs: - lyrical: rtabmap without its tests (MSVC can't build test_optimizer_gtsam.cpp), as jazzy - lyrical: python_qt_binding uses PyQt6 on macOS too (RoboStack/ros-lyrical#86) - rcutils test: Python import only on humble and jazzy Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * lyrical: foxglove_bridge on Windows with /permissive With C++20 (pulled in by lyrical's dependencies) MSVC enables /permissive-, which rejects the Foxglove SDK 0.28's explicit specializations of ParameterView::get after their first use (C2908/C3856 in parameter.hpp); humble/jazzy build it as C++17. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * lyrical: Rust message crates match the rosidl_buffer sequence layout [skip ci] From Lyrical on, rosidl_runtime_c's primitive and string sequences carry two trailing rosidl_buffer flags (ros2/rosidl#942), 32 instead of 24 bytes. rosidl_runtime_rs 0.7 has the matching layout (ros2-rust/rosidl_runtime_rs#22), and rclrs 0.8 uses it, but rosidl_generator_rs 0.5.0 still pins the generated crates to 0.6, so every primitive-sequence field is misread: a Marker or JointState round trip segfaults (RoboStack/ros-lyrical#17, ros2-rust/ros2_rust#659). Pin 0.7. The generated code needs no other change. Checked locally: the generated std_msgs crate pins 0.7, and with 0.7 the visualization_msgs/Marker and sensor_msgs/JointState round trips are correct (rmw Marker 480 bytes, as in C), while 0.6 segfaults. Also point gz_ros2_control's plugin path hook at bin/ on Windows, where the plugin DLL is installed (#104). Evict the generator, every package that generates interfaces (each ships its generated Rust crate) and gz_ros2_control from the PR build cache. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Update mutex_package configuration in vinca.yaml Removed build number and upper bound for mutex_package. * Clear evict_cache in ci.yaml Reset evict_cache to an empty list. * humble and rolling keep Qt5 for the Qt5 rosdep keys This PR maps the Qt5 rosdep keys to Qt6 for every distribution. humble's own override comes with its sync (#62), rolling moves to Qt6 with #99: until then both get the override jazzy has (distros/<d>/robostack.yaml, layered over the shared mapping with conda_index). Their generated recipes (humble linux-64, win-64, osx-arm64; rolling linux-64, win-64) are identical to main's, except for the new rcutils package test, which doesn't change any build hash. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * rosbag2_performance_benchmarking_msgs: no own short-paths flag vinca's Windows build scripts pass -DCMAKE_INTERMEDIATE_DIR_STRATEGY=SHORT for every package since RoboStack/vinca#178 (pinned with #139, which has to be merged before this PR). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Build branches: a newer run cancels the older one A new push to a build branch (after a merge to main) brings the latest recipes and workflow; the older run's plan is out of date. The new run skips what the old one uploaded already, so only the packages its cancelled jobs had built but not uploaded are built again. Re-run failed jobs only in the newest run: re-running an older one would cancel the newer run. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Build times from the jazzy full rebuild (linux-64, linux-aarch64, osx-arm64) Per-recipe build times of the jazzy build-branch runs 39770179894, 39770118028 and 39770103102, from their job logs, so that this sync's lyrical rebuild is already packed by time (osx-64 and win-64 weren't finished yet). The weekly update-build-timings PR adds the times the build jobs record from now on. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Build times for all five platforms, also for lyrical's compatibility recipes osx-64 and win-64 from the jazzy build-branch runs' job logs (39770120710 and 39907983469; 39770136120 and the finished jobs of 39907979075). Recipes those runs didn't reach are estimated from their linux-64 time with a fit per platform (osx-64: 37 s + 1.61 x; win-64: 77 s + 2.13 x; median error 33% and 20%), compatibility recipes at the platform's median (0.3-0.4 s). Every ros-jazzy-<name> time is copied to ros-lyrical-<name> (under "lyrical"), so that lyrical's compatibility recipes have a time until its own build jobs record theirs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test_generate_gha: independent of the repository's build times With shared/build_timings/win-64.json present, rs gha packs by time and first retries an oversized workflow with fewer runners, so the batch-size retry test now runs without timings; a new test covers the retry with timings. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * rs gha: fall back to batching by count when a timed workflow stays too large Packing by time keeps short stages apart (they share a job only up to two minutes), so lyrical's win-64 workflow, whose jobs carry the long Windows scripts, stayed at 621 KiB even with --runners 2. After the runner retries, generate it without --timings; batching by count then merges the short stages and the larger-batch retries shrink it as before. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * vinca with job budgets (RoboStack/vinca#180); rs gha without its own size retries vinca-gha now keeps the workflow under GitHub's size limit itself (a job budget spent where it shortens the workflow most, with or without build times), and its Windows jobs call .scripts/build_win.bat instead of inlining it. rs gha drops the retries with fewer runners and the fallback to batching by count; larger batches remain the last resort when even the fewest jobs the batch size allows don't fit. The check that the Windows workflow uses the repository's build script looks for the call. lyrical's full rebuild: win-64 149 jobs (483 KiB), linux-64 197 jobs (341 KiB). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Tobias Fischer <info@tobiasfischer.info> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
On Windows, long package names push MSVC object file paths past 260 characters.
cl.exethen fails withC1083: Cannot open compiler generated file: '': Invalid argument, andLongPathsEnableddoesn't help. In RoboStack this brokerosbag2_performance_benchmarking_msgs(its generated type support) in the jazzy and lyrical win-64 builds. It is also whyautoware_internal_localization_msgsandiiwa14_moveit_configare excluded on Windows.The Windows build templates (
bld_ament_cmake,bld_catkin,bld_colcon_merge) now pass-DCMAKE_INTERMEDIATE_DIR_STRATEGY=SHORT:CMAKE_INTERMEDIATE_DIR_STRATEGYenvironment variable through to the build script.Tested:
additional_cmake_argsforrosbag2_performance_benchmarking_msgsalone. That package's jazzy win-64 job had failed with C1083 twice in a row. With the flag, the build-branch run built and uploaded it (ros2-rosbag2-performance-benchmarking-msgs-0.26.11-np2py312hdd89193_25), and lyrical's did too.Once this is pinned, RoboStack can drop the per-package setting and retry the packages excluded on Windows for this reason.
🤖 Claude (AI assistant), on behalf of @Tobias-Fischer
🤖 Generated with Claude Code