Re: BUG #19095: Test if function exit() is used fail when linked static

Hi Andres and hackers,
I came up with the solution
— short follow-up with what I changed and how the Meson check actually runs.

Summary of what I did
  - Added a Meson custom_target in src/interfaces/libpq/meson.build that
    scans libpq's object files for direct exit() references.
  - Added pthread_exit to the Makefile whitelist (the Makefile test
    already existed; this just updates the whitelist).

  -Added the custom target in the top level meson.build

Where the Meson check lives and when it runs
  - I put the custom_target in src/interfaces/libpq/meson.build,
    immediately after the libpq shared/static library targets are
    defined and after the declare_dependency for libpq.  That is the
    correct location in the libpq build file.
  - The custom_target declares `depends: [libpq_st, libpq_so]`.  This
    is important: it tells Meson to build those targets first so the
    .o files are present in the build directory before the check runs.
    The check itself does not scan libpq_st or libpq_so; it scans
    files in build/src/interfaces/libpq/*.o.
  - The command uses `find <builddir>/src/interfaces/libpq -name '*.o'`
    and runs `nm -u` on each .o, piping through `grep -v` for the
    whitelisted names and finally checking for exit.
  - The check is skipped for:
      * cross-builds
      * when coverage (b_coverage) is enabled
      * on Windows (no nm/grep in the same form there) as Andrew questioned

Why pthread_exit is whitelisted
  - pthread_exit can legitimately appear in a few build/runtime
    configurations (thread runtimes or link-time glue), and the
    Makefile test was updated to whitelist it.  The Meson check has the
    same whitelist so both build systems behave the same.
  - Whitelisting pthread_exit doesn't remove the value of the test:
    it only avoids false positives for legitimate thread shutdown code.
    We still catch direct exit()/ _exit()/abort()-style calls as tom said.

How to reproduce locally
  - From repo root:
      rm -rf build
      meson setup build
      cd build
      ninja
    The custom_target runs as part of the normal build and will fail
    the build if any .o contains an un-whitelisted exit() reference.

On Fri, Nov 14, 2025 at 7:31 PM Andres Freund <andres@anarazel.de> wrote:

But more generally: If we allow pthread_exit(), what's the point of this test?
That's one of the functions we better avoid calling, no?

 I agree it's worth questioning which functions we allow.  The
 current choice (whitelist pthread_exit) mirrors the Makefile
 behavior, avoids false positives, and keeps the test focused on
 direct process-termination calls authored in our code.  If the
 community prefers a stricter policy so we can adjust the whitelist.

 

ISTM that if we do want to continue having this test, the issue is that we're
testing the shared library - which will have already linked against static
libraries like the sanitizer ones or in this case libcrypto. What we ought to
do is to test the .o files constituting libpq.so, rather than the already
linked .so. That way we will find our own calls to exit etc, but not ones in
static libraries.

 

TBH After so many test runs I have finalized

Why scan .o files (not the final .so)?
  - A shared library is usually linked with other static libraries
    (libcrypto, sanitizer runtimes).  If we scan the final .so we'll
    see references that originate in those static libraries and produce
    false positives.
  - If we scan the .o files that make up libpq, we only inspect our
    own compilation units and will catch only exit() calls introduced
    by our code.

HTH! I attached the patch and also added the meson test output before /after of any file containing an 'exit' explicitly; it fails the build[ninja].

Regards,
Vasuki M
BharatDB[CDAC chennai]