chanced/picker

Utilities that pair with the official Go Elasticsearch package

picker's Issues

Error messages should match those returned from Elasticsearch

Error messages should be exactly what Elastic would return. That way when an error response is parsed from Elastic, errors.Is can function as expected regardless of whether the error originated from picker or elasticsearch.

Convert as many interface{} on the marshaling types to primitive pointer types or dynamic types

In an effort to move things along, I used interface{} for dynamic fields on the internal types used for (un)marshaling.

type completionField struct {
	Analyzer                   string      `json:"analyzer,omitempty"`
	SearchAnalyzer             string      `json:"search_analyzer,omitempty"`
	SearchQuoteAnalyzer        string      `json:"search_quote_analyzer,omitempty"`
	PreserveSeperators         interface{} `json:"preserve_separators,omitempty"`
	PreservePositionIncrements interface{} `json:"preserve_position_increments,omitempty"`
	MaxInputLength             interface{} `json:"max_input_length,omitempty"`
	Type                       FieldType   `json:"type"`
}

These need to be migrated to either primitive pointer types or dynamic types:

type completionField struct {
	Analyzer                   string        `json:"analyzer,omitempty"`
	SearchAnalyzer             string        `json:"search_analyzer,omitempty"`
	SearchQuoteAnalyzer        string        `json:"search_quote_analyzer,omitempty"`
	PreserveSeperators         *dynamic.Bool `json:"preserve_separators,omitempty"`
	PreservePositionIncrements *bool         `json:"preserve_position_increments,omitempty"`
	MaxInputLength             *int          `json:"max_input_length,omitempty"`
	Type                       FieldType     `json:"type"`
}

I'm just not sure which. Things I need to think through:

• If I use primitives, I'll need to add something to the effect of Float64Ptr and CoerceInt64Ptr to the types in the dynamic package. Should I be concerned about the usability of this for the dynamic package? It wont be a pointer to the actual value.
• If I use dynamic, that'll be a lot cleaner but I won't be able to cleanly coerce. This is a problem as some accessors currently coerce int values.

Doing so should improve performance of enc/dec.