mirror of
https://github.com/vel21ripn/nDPI.git
synced 2026-05-19 16:28:13 +00:00
27071bc6e7
Problem: -------- The armhf CI tests were failing because readdir() returns NULL with errno=EOVERFLOW when running 32-bit ARM emulation via QEMU user-mode on a 64-bit host. This is a known QEMU bug (#1805913). Root cause: The host kernel's getdents64 syscall returns 64-bit inode numbers. When QEMU passes these to 32-bit ARM glibc, they overflow the 32-bit range, causing glibc to return EOVERFLOW. This affects any code that lists directory contents (e.g., ndpi_load_categories_dir). See: https://bugs.launchpad.net/qemu/+bug/1805913 https://gitlab.com/qemu-project/qemu/-/issues/263 Possible fixes: --------------- 1. Use tmpfs filesystem - inode numbers start small (1, 2, 3...) and fit in 32-bit range 2. Use real ARM hardware or full system emulation instead of user-mode emulation 3. Skip directory-loading tests on armhf emulation 4. Wait for QEMU to fix the bug (open since 2018) Implemented fix: ---------------- Use tmpfs for the build and test environment: - Switch from 'uses: docker://' to 'run: docker run --privileged' to allow mounting tmpfs inside the container - Create tmpfs at /build and copy entire source tree there - Build and run tests from tmpfs where inode numbers are small This approach was validated locally - all tests pass with tmpfs while they fail on the host-mounted filesystem. Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| build-msbuild.yml | ||
| build-rpm.yml | ||
| build.yml | ||
| build_docker.yml | ||
| build_freebsd.yml | ||
| build_masan.yml | ||
| build_scheduled.yml | ||
| cifuzz.yml | ||
| codeql.yml | ||