The notation of a provider is an independent class that roundtrips

A vendor id is an id scheme, an external id and a field. The description of a quote is a quote_description, today the spot quote of a currency pair and the side of the price (last, bid, ask or mid). A notation is a provider_notation:

The classes are bloomberg_notation, lseg_notation and demo_notation, in ores.marketdata.core/vendor/. The function notation_of(provider) lists them.

• read(write(d)) is d, for every description the notation writes.
• write(read(id)) is id, for every id the notation reads.
• A notation never writes an id that would read as another description. It refuses instead.

A refusal carries one reason: wrong_scheme, wrong_field, malformed, unsupported or not_representable. unsupported means the id is valid for the provider and this description cannot hold it. A Bloomberg ticker with another yellow key, or with a pricing source, is such an id.

The LSEG RIC does not say which currency of a USD pair is the base, so the class holds the market convention: EUR, GBP, AUD and NZD are the base against USD, and every other currency is the quote. It refuses a pair that is not in that order, because JPY/USD would read back as USD/JPY. The LSEG rows are the documented examples, and a person with a login confirms them.

The readers are Boost.Spirit.X3 grammars. X3 is header-only and is already in our Boost. Boost.Parser is not installed, and adding it changes the manifest. The writers are plain code, and the roundtrip tests keep the two together.

• The corpus under projects/ores.marketdata/core/tests/corpus/vendor_notation/ holds the ids of each provider with the description each reads as or the reason it is refused, and the descriptions with the id each provider writes. A script wrote the expectations from the providers' documented examples. The C++ under test did not.
• A grid test writes every ordered pair of twelve currencies on every side through every notation and reads each id back.
• A damage test changes or cuts each character of every valid id. A notation that still reads the damaged id must write the damaged id back, so it never repairs an id quietly.
• A catalogue test reads every installed vendor instrument of these providers and checks that it reads and writes back as itself, so the catalogue and the classes cannot drift apart.

The simulator asks notation_of(provider) for the id of a generator and writes nothing itself. The classes do not depend on the simulator, and the simulator does not hold any provider's grammar.

A rate pillar or another instrument is a further alternative of quote_description. Each notation then reads and writes it or refuses it with unsupported or not_representable. A new provider is a new class and one line in notation_of.