// Summa Query Language Grammar
WHITESPACE = _{ " " | "\t" | "\r" | "\n" }
// Basic tokens
term = @{ (LETTER | NUMBER | "_" | "-" | "." | "'" | "\\" ~ ANY)+ }
quoted_string = @{ "\"" ~ inner_string ~ "\"" }
inner_string = @{ (!("\"" | "\\") ~ ANY | "\\" ~ ANY)* }
number = @{ "-"? ~ ASCII_DIGIT+ ~ ("." ~ ASCII_DIGIT+)? }
// Field specification
field_name = @{ ASCII_ALPHA ~ (ASCII_ALPHANUMERIC | "_")* }
field_spec = { field_name ~ ":" }
// Query types
term_query = { field_spec? ~ !("regex" ~ "(") ~ term }
phrase_query = { field_spec? ~ quoted_string }
// Complete token patterns; consume interior/leading operators as one query.
// An escaped operator is literal; at least one unescaped operator is required.
pattern_literal = _{ "\\" ~ ANY | !(WHITESPACE | "\"" | "(" | ")" | "*" | "?") ~ ANY }
term_pattern = @{ pattern_literal* ~ ("*" | "?") ~ (pattern_literal | "*" | "?")* }
term_pattern_query = { field_spec? ~ term_pattern }
// Explicit whole-term patterns, with JSON string escaping.
regex_query = { field_spec? ~ "regex" ~ "(" ~ quoted_string ~ ")" }
wildcard_query = { field_spec? ~ "wildcard" ~ "(" ~ quoted_string ~ ")" }
// Vector query: field:ann([1.0, 2.0, 3.0], k=10, nprobe=32)
// or field:ann([1.0, 2.0, 3.0], 10)
vector_array = { "[" ~ number ~ ("," ~ number)* ~ "]" }
ann_param = { ("k" | "nprobe" | "rerank") ~ "=" ~ number }
ann_params = { number | ann_param ~ ("," ~ ann_param)* }
ann_query = { field_spec ~ "ann" ~ "(" ~ vector_array ~ ("," ~ ann_params)? ~ ")" }
// Sparse vector query: field:sparse({1: 0.5, 5: 0.3}, k=10)
sparse_entry = { number ~ ":" ~ number }
sparse_map = { "{" ~ sparse_entry ~ ("," ~ sparse_entry)* ~ "}" }
sparse_query = { field_spec ~ "sparse" ~ "(" ~ sparse_map ~ ("," ~ ann_params)? ~ ")" }
// Boolean operators
// Atomic keywords inspect the immediate boundary, without implicit whitespace
// skipping. Do not split identifiers, field names or prefix values (NOTHING,
// NOT:term, NOT*). Symbolic operators do not need a word boundary.
keyword_end = _{ &(WHITESPACE | "(" | "\"" | EOI) }
and_op = @{ "AND" ~ keyword_end | "&&" }
or_op = @{ "OR" ~ keyword_end | "||" }
not_op = @{ "NOT" ~ keyword_end | "!" }
// Modifiers bind to the immediately following operand; a bare `+`/`-`
// followed by whitespace is ordinary text.
required_op = @{ "+" ~ !WHITESPACE }
prohibited_op = @{ "-" ~ !WHITESPACE }
// Grouping
group = { "(" ~ or_expr ~ ")" }
// Primary expression (vector queries added)
primary = { (required_op | prohibited_op | not_op)? ~ !("+" | "-" | "!") ~ (group | ann_query | sparse_query | regex_query | wildcard_query | phrase_query | term_pattern_query | term_query) }
// Boolean expressions
and_expr = { primary ~ (and_op ~ primary)* }
// Implicit OR: space-separated and_exprs without an explicit OR operator
// are treated as OR (matching parse_plain_text semantics).
or_expr = { and_expr ~ (or_op ~ and_expr | and_expr)* }
// Top-level query
query = { SOI ~ or_expr ~ EOI }