Changelog

Source: NEWS.md

databentoR 0.1.0

First release. A complete R client for the Databento HTTP API, mirroring the official Python client 0.86.0 on the wire.

Historical API

• db_get_range() streams a range and returns a tibble, optionally writing parquet at the same time. It requests the CSV encoding rather than binary DBN, so no decoder is needed, and it is the only request the package never retries, because a retried stream can be billed twice.
• Metadata, all free of charge: db_list_publishers(), db_list_datasets(), db_list_schemas(), db_list_fields(), db_list_unit_prices(), db_get_dataset_condition(), db_get_dataset_range(), db_get_record_count(), db_get_billable_size() and db_get_cost().
• db_resolve() resolves symbols and returns one row per validity interval, with unresolved symbols attached rather than dropped.
• Batch jobs: db_batch_submit_job(), db_batch_get_job_details(), db_batch_list_jobs(), db_batch_list_files() and db_batch_download(), with checksum verification and skip-if-complete restarts.

Reference API

• db_adjustment_factors(), db_corporate_actions(), db_corporate_action_events(), db_corporate_action_enums(), db_security_master() and db_security_master_last().

Correctness

• Column types are assigned from the field name, never inferred. arrow’s CSV inference reads a trades slice whose action column is all "T" as boolean, which would silently corrupt every trade action. db_field_types() reports what will be assigned, and takes a schema argument because exactly one field name is schema-dependent: action is a character code in the trade and book schemas and a numeric enum in status.
• ts_type = "integer64" returns exact nanosecond timestamps. The default POSIXct stores seconds as a double and resolves to about a quarter of a microsecond on modern dates.
• Fields that are 64-bit on the wire come back as bit64::integer64 instead of being downcast to double, so order_id, raw_instrument_id and an undefined statistics quantity keep their value. The last of those is the int64 maximum, which a double rounds.
• Symbols are normalised as the Python client normalises them: upper-cased, with the roll rule of a continuous symbol lower-cased, NULL becoming ALL_SYMBOLS, and the documented 2000-symbol cap enforced before the request is spent.
• Date parameters reject timestamps, as in the Python client, rather than silently truncating an instant to a day.
• HTTP 206 raises a warning: it is a success code that means only some symbols were resolved.
• The X-Warning response header is surfaced as an R warning, so server-side deprecation notices are not swallowed.

Equivalence suite

• Wire equivalence against the pinned Python client for all 28 endpoints, offline and free, asserting method, endpoint, and parameter names, values and order. The R value domains are checked against the compiled DBN enums.
• Data equivalence over six tiny slices spanning ohlcv-1d, trades, tbbo, mbp-1, statistics and definition, with prices compared to a relative tolerance of 1e-12. Both sides are sorted before comparison: the server does not promise a stable order among records sharing a timestamp, and neither client can.
• A full live coverage sweep over every exported function: all twenty schemas across three datasets, a batch job submitted, polled, downloaded and compared against the streamed equivalent, symbology for every input symbol type, and the reference endpoints. It runs weekly and on demand, and never on a push.
• A weekly workflow diffs a new upstream release against the pin and files the changed calls as an issue.

Downloaded market data is never committed; the reference fixtures stay untracked.