Function Identity and Observability Specification

Table of Contents

• 1. Problem to Measure
• 2. Recorded Direction
• 3. Explicit Boundaries
• 4. Future Plan

This deferred specification records the possible optimization of function-name filtering and profiling. It is not part of the field-statistics implementation and currently has no implementation branch.

1. Problem to Measure

Logging allowlists and profiling currently identify instrumented functions by name. Emitted logging is normally dominated by formatting and I/O, but filtered calls in hot loops and repeated profiler start/end name searches may have measurable cost. That cost must be benchmarked before changing the system.

2. Recorded Direction

If benchmarks justify the work:

• give instrumented application functions a compile-time FunctionId or a call-site-cached registry handle;
• resolve configured names once into a bitset/indexed allow table;
• make profiler start/end update directly indexed state;
• check log level before function filtering in every hot-path macro;
• parse profiling mode once at ingress; and
• retain canonical function names for YAML, diagnostics, errors, and output.

The implementation should preserve existing logging macros, log formats, profiling files, and default behavior as far as possible. A cached handle may be less intrusive than a manually maintained enum of every C function; the benchmark/prototype phase chooses between them.

3. Explicit Boundaries

This work does not:

• reuse FieldId or create one universal identity namespace;
• assign IDs to filenames, arbitrary configuration values, PETSc object names, DMSwarm field names, or user-defined postprocessing labels;
• replace boundary-condition enums/function pointers already resolved at ingress;
• redesign runtime physical-solution monitoring or scientific statistics; or
• block Field Statistics.

One-time string comparisons at external ingress and presentation boundaries are correct and remain.

4. Future Plan

1. Benchmark disabled, filtered, selected, and all-function logging/profiling on representative serial and MPI runs.
2. Inventory call sites and macros; identify only repeated runtime lookups.
3. Prototype cached handles and compile-time IDs without changing output.
4. Select the simpler design that demonstrates a material improvement.
5. Add behavior, concurrency, and overhead regressions, then update monitoring and contributor documentation.

This work receives its own branch series only after explicit approval.