Package 'rcutils' exports the library 'rcutils' which couldn't be found

[lesomnus]

Generated by Generative AI

No response

Operating System:

build arm64 ros:humble docker image on amd64 machine with buildkit

ROS version or commit hash:

ros:humble 8a99e0ca6f90

RMW implementation (if applicable):

No response

RMW Configuration (if applicable):

No response

Client library (if applicable):

No response

'ros2 doctor --report' output

ros2 doctor --report

<COPY OUTPUT HERE>

Steps to reproduce issue

colcon build

Expected behavior

succeed to build

Actual behavior

#19 14.28 CMake Error at /opt/ros/humble/share/rcutils/cmake/ament_cmake_export_libraries-extras.cmake:48 (message):
#19 14.28   Package 'rcutils' exports the library 'rcutils' which couldn't be found
#19 14.28 Call Stack (most recent call first):

Additional information

There is no problmen on amd64

[frapit]

Happens to me as well on arm runners and is probably caused by the latest commit 5dd09d2 since it works if I go back to b11330c.

[mjcarroll]

@fujitatomoya this looks like it came from #516 (or the backport in #519). Do you know what is happening here?

[mjcarroll]

My understanding is that this shouldn't return true on arm64:

check_library_exists(atomic __atomic_load_8 "" HAVE_LIBATOMICS)

[fujitatomoya]

@mjcarroll thanks for the ping.

this is weird... we have(had) CI aarch 64 no problem, e.g https://ci.ros2.org/job/ci_linux-aarch64/20157/. if this is true this should fail all the time?

[fujitatomoya]

@frapit @lesomnus are you seeing this problem with arm64 ros:humble docker image only, what if you build the rcutils on the arm64 host system? sorry, i do not have arm64 host now.

[fujitatomoya]

how about #528 (draft)?

[mjcarroll]

this is weird... we have(had) CI aarch 64 no problem, e.g https://ci.ros2.org/job/ci_linux-aarch64/20157/. if this is true this should fail all the time?

Good point. Everything I could find around that missing symbol seemed to be aarch64 related, though.

[rgov]

I do not believe this is related to #516. It is unclear why that change would cause a non-deterministic build failure, and why it seems to affect some arm64 builds (often relying on emulation), but not others like the ROS2 Jenkins CI jobs.

After burning an evening on this, I determined the builds where rcutils resolves and those where it doesn't have identical traces, modulo the naming of random intermediate files (cmTC_[0-9a-f]{5} and cc[A-Za-z0-9]{6}).

I tried reverting back to QEMU 7.0 (~April 2022) but it did not resolve the issue.

[rgov]

I instrumented CMake with strace to see where the behavior diverges. In both cases, the directory listing syscall getdents64() on /opt/ros/humble/lib returns the same number entries.

In the failure case, there's a failed mmap() call around the same time:

  mmap(NULL, 36864, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f2832398000
  getdents64(6, 0x7f2832398870 /* 256 entries = [... {d_ino=4543995, d_off=95, d_reclen=40, d_type=DT_REG, d_name="librcutils.so"}...] /*, 32768) = 16032*
  munmap(0x7f2832398000, 36864)           = 0

>> some other allocation happens in the failure case
  mmap(0x7f28379ed000, 143360, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED_NOREPLACE, -1, 0) = -1 EEXIST (File exists)
  mmap(NULL, 143360, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f283237e000

  mmap(NULL, 36864, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f2832375000
  getdents64(6, 0x7f2832375880 /* 0 entries = [] */, 32768) = 0
  munmap(0x7f2832375000, 36864)           = 0

>> failure case
  write(1, "-- DEBUG: find_library rcutils _"..., 50) = 50
  write(2, "CMake Error at /opt/ros/humble/s"..., 1182) = 1182

>> success case
  close(6)                                = 0
  faccessat(AT_FDCWD, "/opt/ros/humble/lib/librcutils.so", R_OK) = 0
  access("/usr/gnemul/qemu-aarch64/opt/ros/humble/lib/librcutils.so", F_OK) = -1 ENOENT (No such file or directory)
  stat("/opt/ros/humble/lib/librcutils.so", {st_mode=S_IFREG|0644, st_size=93360, ...}) = 0
  write(1, "-- DEBUG: find_library rcutils _"..., 70) = 70
  faccessat(AT_FDCWD, "/opt/ros/humble/lib/librcutils.so", R_OK) = 0