Repository navigation
Upload and cache noarch packages - #212
Draft
EzraBrooks wants to merge 5 commits into
Draft
EzraBrooks wants to merge 5 commits into
EzraBrooks wants to merge 5 commits into
Conversation
Drop when lyrical-python-abi has merged: git rebase --onto upstream/main <this commit> py-rattler
Along the way, fixed a bug with `~=` version constraints and added a new suite of tests for version ordering and constraints.
vinca builds the ros-<distro>-<name> compatibility packages as noarch: generic. Every platform's build uploads output/noarch too (anaconda.org is checked again before each upload attempt, so an upload lost to another platform's job counts as done), the pull-request cache keeps it, and the gap report counts noarch packages for every platform.
The vinca PR's commit on EzraBrooks/vinca, until it merges. It is based on vinca's master after the py-rattler changes (vinca#196), which this branch already builds on. Its get_channel_records reads the channel's noarch subdir too, so the package tests and the released build numbers no longer read noarch separately (the package tests would select every noarch package twice).
Tobias-Fischer
added a commit
to Tobias-Fischer/robostack-fork
that referenced
this pull request
Nov 13, 2026
…ages vinca-rs builds them as noarch: generic (Tobias-Fischer/vinca-rs#5, from RoboStack/vinca#190 and RoboStack#212): every platform's jobs build and upload them, the upload skips what's already there, and generate's skip_existing reads the channel's noarch subdir. Takes effect with the vinca-rs release that has it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tobias-Fischer
added a commit
that referenced
this pull request
Nov 18, 2026
Reverts 5bd21ed: humble uses the shared ros2-mrpt-common.patch again. The bare -DCMAKE_AR=<triplet>-ar (and CMAKE_STRIP, ...) in conda's CMAKE_ARGS came from conda-forge's ctng-compiler-activation (#209, gcc_linux-* 15.3.0 build 3, published 2026-11-17); #212 restored the absolute paths (build 4, 2026-11-18). With build 4, mrpt_common/mrpt_core (pybind11 strip) and mrpt_maps (static cloudini) build on linux-aarch64 without it. The same regression broke libnabo's static library in CI run 44064052374. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tobias-Fischer
added a commit
that referenced
this pull request
Nov 18, 2026
Reverts 4791e5f and 5cd0682: jazzy uses the shared ros2-mrpt-common.patch again. The bare -DCMAKE_AR=<triplet>-ar (and CMAKE_STRIP, ...) in conda's CMAKE_ARGS came from conda-forge's ctng-compiler-activation #209 (gcc_linux-* 15.3.0 build 3); #212 restored the absolute paths (build 4). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tobias-Fischer
added a commit
that referenced
this pull request
Nov 20, 2026
* jazzy: rosdistro sync 2026-11-16 * jazzy: drop ogre from gz_rendering_vendor conda-forge's gz-rendering8 no longer links OGRE 1 (8.2.2_1), so the unpinned ogre that gz_rendering_vendor's libogre-dev key maps to resolved to 14.6 and its run_export clashed with gz-gui8 (still built against ogre 1.10) in gz_gui_vendor's host environment: every jazzy build job of the 2026-11-16 sync stopped there. libogre-dev is only needed to build gz-rendering from source. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: mrpt_common: find the strip program for pybind11 modules MRPT 3.6.2 packages find pybind11 (23 of them have it in host) and build their Python bindings. pybind11 strips the module with CMAKE_STRIP, which conda's CMAKE_ARGS pass as a bare program name and CMake turns into a path in the build directory, so linking failed with "<triplet>-strip: not found" (mrpt_core on linux-aarch64). mrpt_add_python_module now resolves it on PATH. The shared ros2-mrpt-common.patch plus this hunk; it can move to shared/ once the other distributions' MRPT needs it too. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: aruco_ros: use the shared patch again aruco_ros 5.0.6 has upstream's tf2 .hpp includes and target_link_libraries; only the opencv4/ include remains, which is exactly shared/patch/ros2-aruco-ros.patch. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: foxglove_bridge: refresh patch for 3.6.3 (SDK 0.30.2) Same change as before (prebuilt SDK for macOS and Windows, <map> include) with the SHA-256 digests of the foxglove-sdk 0.30.2 release assets. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: mola_launcher: patches for 3.4.0 3.4.0 reworked the module loader, so the shared Windows patch no longer applies; jazzy gets its own copies (they replace the shared ones per package). The generic patch is the shared one plus dlerror() on macOS too (3.4.0's new error path uses GetLastError() for anything but __unix__); the Windows patch is ported to the new loader (LoadLibraryA into void*, GetLastError() cast). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: popf: refresh macOS patch for 1.0.1 Same changes on 1.0.1's CMakeLists.txt (C++14, no GNU ld flags on Apple, FindOSI/BISON/FLEX before include_directories, FLEX include dir). Dropped the hunks for files 1.0.1 no longer builds (VALfiles/parsing/CMakeLists.txt, fixyywrap, test/lex.yy.cc, the unused ParserGlobals.cpp) and the FindOSI one (upstream now also looks for coin/OsiConfig.h). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: sick_scan_xd: drop macOS patch, fixed upstream 3.10.0 no longer passes the GCC-only flags on Apple. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: mrpt_common: resolve every bare binutils name, not just strip CMAKE_AR is affected the same way as CMAKE_STRIP: mrpt_maps 3.6.2 bundles cloudini as a static library, whose archive step ran $SRC_DIR/build/<triplet>-ar (not found) on linux-aarch64. The check moves from mrpt_add_python_module to the top of mrpt_cmake_functions.cmake and covers AR, RANLIB, STRIP, NM, OBJCOPY, OBJDUMP, READELF and ADDR2LINE; it does nothing where the configured path exists. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: evict the gz_rendering_vendor built before the ogre fix from the PR cache Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: mrpt_common: drop the binutils workaround again Reverts 4791e5f and 5cd0682: jazzy uses the shared ros2-mrpt-common.patch again. The bare -DCMAKE_AR=<triplet>-ar (and CMAKE_STRIP, ...) in conda's CMAKE_ARGS came from conda-forge's ctng-compiler-activation #209 (gcc_linux-* 15.3.0 build 3); #212 restored the absolute paths (build 4). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: plansys2_pddl_parser: allow floating-point std::from_chars on macOS Same as humble (e76b226): 4.0.0's Utils.cpp uses floating-point std::from_chars, which libc++ marks unavailable on macOS; conda-forge's libc++ has it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: opennav_docking_bt: build without -Werror gcc 15 warns -Wnull-dereference in behaviortree_cpp's thread_safe_signal.h (linux-aarch64). nav2_common's -Werror is already patched out (ros2-nav2-common.patch); this package sets its own flags. Unix only, as the Windows patch rewrites the same line. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: realsense2_camera: accept conda-forge's librealsense 2.58.3 4.59.1 asks for realsense2 2.59.0, but librealsense2 is conda-forge's 2.58.3 (override_version in pkg_additional_info.yaml, until librealsense-feedstock#91). Same as rolling's patch. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: plansys2_core, plansys2_popf_plan_solver: allow floating-point std::from_chars on macOS As for plansys2_pddl_parser: 4.0.0 parses doubles with std::from_chars, which libc++ marks unavailable on macOS; conda-forge's libc++ has it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * mrpt_imgui: build with GDI on Windows, as mrpt_gui portable-file-dialogs.h includes commdlg.h, which needs wingdi.h's LOGFONT; the ROS build scripts set CL=/DNOGDI=1 (C3646 on 'lpLogFont' on win-64 in jazzy and humble). Only mrpt_imgui's Windows recipe changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: plansys2_core: create the planner pipe without pipe2 on macOS PlanSolverBase.cpp uses pipe2(O_CLOEXEC), which the macOS SDK in CI lacks (undeclared identifier); pipe() plus FD_CLOEXEC on both ends. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: ur_controllers: link the twist_controller parameters 3.10.0 generates twist_controller_parameters but doesn't link it, so twist_controller.cpp can compile before its header exists (failed on linux-aarch64: twist_controller_parameters.hpp: No such file). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: kairos: allow floating-point std::from_chars on macOS lexer.cpp parses numbers with std::from_chars, which libc++ marks unavailable on macOS; conda-forge's libc++ has it (as for plansys2). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * mp2p_icp_core: no C++ module scanning for mp2p_icp_filters on Windows CMake 4.5 builds the std module for the C++20 mp2p_icp_filters target, which inherits mola_common's /std:c++latest and a /std:c++17 that overrides it: yvals_core.h C1189 'Standard Library Modules are available only with C++20' (jazzy win-64, 3.1.0). No modules are used. Shared: 3.0.2 (humble, rolling) has the same target and gets the same CMake. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: toppra: package 0.6.10, which conda-forge's roboplan is built against ros2-toppra followed the snapshot's 0.6.11 (toppra-python >=0.6.11), but roboplan-toppra-python 0.8.0 needs libtoppra 0.6.10: no roboplan_* recipe could resolve its host environment on linux. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: plansys2_popf_plan_solver: include unistd.h for mkdtemp on macOS popf_plan_solver.cpp calls mkdtemp() without <unistd.h>, where macOS declares it (undeclared identifier on osx-arm64). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: plansys2_kairos_plan_solver: include unistd.h for mkdtemp on macOS As for plansys2_popf_plan_solver: kairos_plan_solver.cpp calls mkdtemp() without <unistd.h>, where macOS declares it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: turtlebot3_node: _USE_MATH_DEFINES on Windows 2.3.7's odometry.cpp uses M_PI (C2065 on win-64). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ros2_pulse: st_mtimespec on macOS 0.7.0's pulse_bridge.cpp reads struct stat's Linux-only st_mtim (jazzy and humble osx-arm64). Shared: only humble and jazzy package ros2_pulse. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: plansys2_bringup: pthread_setschedparam on macOS 4.0.0's plansys2_node calls sched_setscheduler(), which macOS lacks; set the main thread's policy instead (as lyrical's patch does). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: toppra: back to the snapshot's 0.6.11 Undoes d3d8c72: conda-forge rebuilt roboplan 0.8.0 against libtoppra 0.6.11 (conda-forge/roboplan-feedstock#28, build 1 on all platforms). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: plansys2, kairos: _LIBCPP_DISABLE_AVAILABILITY as a CMake flag instead of patches pkg_additional_info.yaml passes it on macOS (CMAKE_CXX_FLAGS_INIT, appended to conda's CXXFLAGS), so plansys2_pddl_parser's and kairos's patches go and plansys2_core's and plansys2_popf_plan_solver's keep only their pipe2 and mkdtemp hunks. Verified locally: both build at macOS 11.0 without the patches. ci.yaml evicts the four from the PR cache so CI rebuilds them (temporary). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: evict the plansys2 packages and kairos instead of gz (temporary) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * jazzy: reset evict_cache The plansys2 packages and kairos were rebuilt with the CMake flag (run 44635453194). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Tobias-Fischer <5497832+Tobias-Fischer@users.noreply.github.com> Co-authored-by: Tobias Fischer <info@tobiasfischer.info> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Tobias-Fischer
added a commit
that referenced
this pull request
Nov 20, 2026
* humble: rosdistro sync 2026-11-16
* humble: refresh the patches the 2026-11-16 sync broke
- rcutils 5.1.11: drop the __EMSCRIPTEN__ hunks of shared_library.c, upstream
has them now
- foxglove_bridge 3.6.3: the SDK is 0.30.2; new SHA-256s of its macOS and
Windows archives (from the GitHub release's digests, as in jazzy)
- popf 1.0.1: upstream now finds OSI/flex, has VALGlobals.cpp and links the
solver libraries itself; keep C++14, the platform-guarded --no-undefined,
and the libc++/values.h/unistd.h fixes (as jazzy's refreshed patch)
- mola_launcher 3.4.0: humble's own copy of the shared patches (as jazzy has),
refreshed for the new MolaDLL_Loader.cpp; the shared ones match the other
distributions' older mola
- libmavconn 2.17.0 (win), mavlink 2026.11.11: refreshed for new context
(cmake_minimum_required 3.10, `defined(__APPLE__) && __APPLE__`)
- sick_scan_xd 3.10.0: drop the osx patch, upstream skips those GCC flags on
Apple now
- plansys2_tools 4.0.0: drop the Qt6 patch, upstream follows qt_gui_cpp's Qt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: mrpt_common: resolve bare binutils names (as jazzy)
humble has the same MRPT 3.6.2 as jazzy, whose linux-aarch64 builds failed
because conda's bare CMAKE_AR/CMAKE_STRIP names end up cached as paths in the
build directory (bundled cloudini in mrpt_maps, pybind11's strip step). Same
patch as distros/jazzy/patch/ros2-mrpt-common.patch: the shared patch plus a
PATH lookup of those tools.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2_pddl_parser: allow floating-point std::from_chars on macOS
4.0.0's Utils.cpp parses numbers with floating-point std::from_chars, which
libc++'s availability markup rejects on macOS ('from_chars' is unavailable,
osx-arm64 CI). conda-forge's libc++ (run-exported, >= 21) has it, so define
_LIBCPP_DISABLE_AVAILABILITY, as shared/patch/ros2-ros2-pulse.osx.patch does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: mrpt_common: drop the binutils workaround again
Reverts 5bd21ed: humble uses the shared ros2-mrpt-common.patch again. The bare
-DCMAKE_AR=<triplet>-ar (and CMAKE_STRIP, ...) in conda's CMAKE_ARGS came from
conda-forge's ctng-compiler-activation (#209, gcc_linux-* 15.3.0 build 3,
published 2026-11-17); #212 restored the absolute paths (build 4, 2026-11-18).
With build 4, mrpt_common/mrpt_core (pybind11 strip) and mrpt_maps (static
cloudini) build on linux-aarch64 without it. The same regression broke
libnabo's static library in CI run 44064052374.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: realsense2_camera: accept conda-forge's librealsense 2.58.3
4.59.1 asks for realsense2 2.59.0, but librealsense2 is conda-forge's 2.58.3
(override_version in pkg_additional_info.yaml, until
librealsense-feedstock#91). Same as rolling's patch.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2_core, plansys2_popf_plan_solver: allow floating-point std::from_chars on macOS
As for plansys2_pddl_parser: 4.0.0 parses doubles with std::from_chars,
which libc++ marks unavailable on macOS; conda-forge's libc++ has it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2_core: create the planner pipe without pipe2 on macOS
PlanSolverBase.cpp uses pipe2(O_CLOEXEC), which the macOS SDK in CI lacks
(undeclared identifier); pipe() plus FD_CLOEXEC on both ends.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2_tools: use Qt6, as qt_gui_cpp
4.0.0 follows qt_gui_cpp_USE_QT_MAJOR_VERSION, which humble's qt_gui_cpp
extras set to 5 although RoboStack builds it with Qt6
(ros2-qt-gui-cpp.patch): Qt5 and Qt6 targets mixed (INTERFACE_QT_MAJOR_VERSION
error on linux).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: libmavconn: no sched.h on Windows
2.17.0 includes <sched.h> in thread_utils.hpp (C1083 on win-64); its only
use, set_this_thread_sched_fifo(), is Linux-only already. Guard it with
pthread.h.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: kairos: allow floating-point std::from_chars on macOS
lexer.cpp parses numbers with std::from_chars, which libc++ marks
unavailable on macOS; conda-forge's libc++ has it (as for plansys2).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2_popf_plan_solver: include unistd.h for mkdtemp on macOS
popf_plan_solver.cpp calls mkdtemp() without <unistd.h>, where macOS
declares it (undeclared identifier on osx-arm64).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2_kairos_plan_solver: include unistd.h for mkdtemp on macOS
As for plansys2_popf_plan_solver: kairos_plan_solver.cpp calls mkdtemp()
without <unistd.h>, where macOS declares it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: mola_launcher: same patches as jazzy (same 3.4.0)
humble's refreshed patch missed the hunk for the error-message branch, so
macOS compiled the Windows GetLastError() path (osx-arm64).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2_bringup: pthread_setschedparam on macOS
As for jazzy: 4.0.0's plansys2_node calls sched_setscheduler(), which macOS
lacks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: turtlebot3_node: _USE_MATH_DEFINES on Windows
2.3.6's odometry.cpp uses M_PI too (C2065 on win-64), as jazzy's 2.3.7.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: plansys2, kairos: _LIBCPP_DISABLE_AVAILABILITY as a CMake flag instead of patches
pkg_additional_info.yaml passes it on macOS (CMAKE_CXX_FLAGS_INIT, appended to
conda's CXXFLAGS), so plansys2_pddl_parser's and kairos's patches go and
plansys2_core's and plansys2_popf_plan_solver's keep only their pipe2 and
mkdtemp hunks. Verified locally: both build at macOS 11.0 without the patches.
ci.yaml evicts the four from the PR cache so CI rebuilds them (temporary).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* humble: reset evict_cache
The plansys2 packages and kairos were rebuilt with the CMake flag (run 44638534200).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Tobias-Fischer <5497832+Tobias-Fischer@users.noreply.github.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.
Companion PR to RoboStack/vinca#190.
Based on #211, #193.