Limits and count

Low-level defaults are:

query.Limits{
    MaxQueryBytes:      16 << 10,
    MaxFilterBytes:     8 << 10,
    MaxTokens:          512,
    MaxLiteralBytes:    4 << 10,
    MaxInValues:        100,
    MaxLimit:           100,
    MaxOffset:          100_000,
    MaxSortTerms:       5,
    MaxSearchBytes:     256,
    MaxExpressionDepth: 16,
    MaxNodes:           100,
    MaxPathDepth:       8,
    MaxQuantifierDepth: 4,
    MaxCursorBytes:     4 << 10,
}

All custom values passed to WithLimits are positive. Engine additionally requires a positive DefaultLimit, MaxLimit, and MaxOffset; an omitted list limit can therefore never produce an unbounded query.

Decoded query bytes include parameter names and values, even for unknown parameters. Filter bytes are rejected before lexing; token and literal bounds then constrain parser allocations and work. Each in or not in list is independently bounded by MaxInValues. Cursor bytes are rejected before base64url/JSON decoding.

List fetches one extra row to calculate has_more. Count is disabled unless AllowCount is true and runs only for count=true. It includes base scope, filter, and search while excluding query sort, limit, and offset. It also excludes cursor position, so every continuation can report the same complete filtered total. Relationship filters use correlated subqueries, so matching child rows do not multiply the root count.

limit=0 executes no data query but may still execute count. Most endpoints should rely on has_more and enable exact count only when clients need it. When has_more is true, nextCursor continues after the last returned item.

Stable benchmark baseline

On the published local reference host (AMD Ryzen 5 2600, Linux amd64, Go 1.26.5), representative v1.0 results were:

Operation	ns/op	B/op	allocs/op
parse HTTP query	2,567	1,744	13
parse maximum valid filter	53,611	76,608	158
parse + apply engine plan	9,215	6,368	64
relationship parse + apply	15,533	9,135	83
cursor parse + apply	21,693	14,106	102
SQLite list	345,911	31,489	699

Benchmarks are regression checks, not latency promises. The full command, environment, and additional cases are in the benchmark record.

Stable benchmark baseline​

Stable benchmark baseline