Also query name.* fields when street is present

[fcruzel]

Bug description

Right now Pelias only queries address_parts.* fields (e.g., address_parts.street, address_parts.number) when a street is present, and completely omits name.* phrase queries in those scenarios. This is missing results, since for example importing alternative names (think name_json though csv-importer) gets put into the name.default route, and that is not searched in the instances where street is detected in the input.

Steps to reproduce

• Make sure to enable libpostal
• Search an address that libpostal is able to detect a street within. E.g.: "ZULOAGA 1 SAN BARTOLOME DE TIRAJANA", gets parsed as

  "parser": "libpostal",
  "parsed_text": {
      "street": "zuloaga",
      "housenumber": "1",
      "city": "san bartolome de tirajana"
  }

• This should be enough for Pelias to only search in address_parts, ignoring name.* routes.

[missinglink]

Please provide examples of the expected behaviour so I can better understand the use case.

[fcruzel]

We use csv-importer to import address records. For each address we populate name_json with a list of alternative spellings/typos of the canonical street name (e.g. OCR artifacts, common misspellings). These end up stored in name.default as an array. address_parts.street only contains the canonical name. For example, we have this record in elastic:

{
  "name": {
    "default": [
      "CALLE TEOBALDO POWER 49 LAS PALMAS DE GRAN CANARIA",
      "CALLE TOBALDO POWER 49 LAS PALMAS DE GRAN CANARIA",
      ...
    ]
  },
  "address_parts": {
    "street": "CALLE TEOBALDO POWER",
    "number": "49"
  }
}

Consider this query: CALLE TOBALDO POWER 49 LAS PALMAS DE GRAN CANARIA

libpostal detects a street in the input, so Pelias builds a query against address_parts.street / address_parts.number only. Since address_parts.street only contains the canonical spelling and the match is exact, the document is not returned.

I expect that if a street is detected, Pelias should also include a name.* clause in the query (alongside the existing address_parts.* clauses), so that records whose alternative names match the input are still surfaced.

I think this is essentially a should clause on name.default in addition to the must clauses on address_parts.*. Documents that satisfy the address_parts path would still rank higher, but records that only match via name.default would not be silently dropped as valid answers.