Developer reports
glibc loader search ledger
Organize observed library-search attempts without inferring which candidate succeeded.
Your inputs
Start with an exampleYour result
Example resultHow it works
Keep separate lookup contexts by PID. Link explicit search-path/cache sources and trying-file rows to the current observed lookup. Preserve dependency requests, mapping starts, initializer entries and unknown/orphan rows separately.
Two paths are attempted under RUNPATH. A later generating-link-map row and calling-init row are separate observations; neither selects the last candidate or proves completion.
Limits
Pinned glibc 2.42 templates, up to 20000 rows, 1000 lookups and 10000 attempts. Raw rows remain; space-normalized PID separators are marked. Ambiguous token forms and other categories stay unknown. No process execution, namespace guessing for init, selected-path/failure diagnosis, constructor completion or process-success verdict.