How do I fix valgrind errors?

How do I fix a valgrind failure on the nightly or on CDash?

1. Classify the failure on the CDash dynamic analysis page. The page is per build and per component, e.g. https://my.cdash.org/builds/<buildid>/dynamic_analysis/<component>. The memcheck configuration lives in CTest.cmake: CTEST_MEMORYCHECK_COMMAND_OPTIONS and CTEST_MEMORYCHECK_SUPPRESSIONS_FILE.
2. Fix real bugs in code. The following are defects in our code, and must be fixed, not suppressed:
   • Invalid read or write of size N.
   • Use of an uninitialised value.
   • Definitely lost or indirectly lost blocks.
   • Mismatched free or delete.
3. Suppress third-party noise. A block reported as still reachable (or possibly lost from a library's one-time global init) is not a leak in our code: the pointer is still reachable in the library's own state, and the library frees it only at process exit (or via its cleanup API, e.g. nats_library_cleanup for the NATS C client). Suppress it in build/valgrind/custom.supp:
   • One entry per allocation site.
   • match-leak-kinds: reachable (add possible only when the record needs it).
   • List the fun: frames from the allocating function up to the library's init entry point; ... matches any middle frames.
   • Add a comment naming the library and the cause, e.g. the NATS C library's nats_openLib one-time init (threads, mutexes, conditions) reached from the first client::connect().
   • Keep entries narrow. A broad suppression hides real bugs; see the sprint_02 Add Valgrind suppressions task.

4. Re-run under valgrind with the same options the nightly uses, pointing at the suppression file, until ERROR SUMMARY is clean:

   valgrind --trace-children=yes --quiet --tool=memcheck \
       --leak-check=full --show-reachable=yes --num-callers=50 \
       --demangle=yes --gen-suppressions=all \
       --suppressions=build/valgrind/custom.supp <test-binary>
   

5. Confirm on CDash on the next nightly run. The workflows that submit dynamic analysis run on schedule, not on PRs.

build/valgrind/custom.supp — the suppression file, committed and shipped with the repo. CTest.cmake — the nightly memcheck invocation and suppression wiring. The failing tests and their valgrind output appear on CDash under Dynamic Analysis for the build.

The three components that failed on the 2026-08-10 nightly (ores.trading.core.tests, ores.reporting.core.tests, ores.refdata.core.tests) ran clean under valgrind with the suppression file after the hotfix.

• Add Valgrind suppressions — the sprint_02 task that created the suppression file and its narrowness rule.