Format parse comparison #

This non-gating harness compares the JSON, INI, and TOML integrations using the same 21 string values grouped into five logical sections. It measures in-memory parse, conversion to a config.Document, and destruction of every owner. File I/O is intentionally excluded.

Run it from the odin_config repository with the pinned Reference Odin compiler:

odin run benchmarks -o:speed -vet -vet-style -warnings-as-errors -collection:external=external

The driver performs one warmup followed by seven batches of 50,000 operations per format. In addition to the complete-operation comparison, it attributes the TOML path across semantic parsing, conversion into config.Document, TOML DOM destruction, and config destruction. The conversion phase includes config.New and candidate commit but excludes destruction. Each phase otherwise times only its named operation; required setup and cleanup run outside that phase's accumulated duration. Timer reads are therefore included in every phase operation, and the sum is an approximation rather than an identity with the batched end-to-end result.

The conversion phase mirrors the current config_toml conversion in benchmark code so it can reuse one parsed document. The end-to-end config_toml.Parse measurement remains authoritative if the adapter changes. The harness reports allocator requests for complete operations and separately for TOML semantic parse and config conversion. Requested-byte totals reflect each integration's arena strategy and are not peak-live-byte measurements.

The JSON and TOML encodings contain the same logical keys and string values but are not byte-identical: the JSON fixture is 509 bytes and the TOML fixture is 478 bytes. INI is retained as useful local context, not as a semantically identical format comparison.