Matches Diesel for type-level filters

Performance

Prax is highly optimized for performance, matching Diesel for type-level operations while providing a developer-friendly Prisma-like API.

Key Achievements

5.1ns

Type-level AND(5)

Matches Diesel!

3.8ns

IN(10) SQL generation

5.8x faster with patterns

64B

Filter enum size

Fits in single cache line

30%

Faster than SQLx

Database execution

Query Building Performance

Operation	Prax	Diesel	SQLx	Notes
Simple SELECT	40ns	278ns	5ns	7x faster than Diesel
SELECT + filters	105ns	633ns	5ns	6x faster than Diesel
INSERT query	81ns	-	5ns	-
UPDATE query	101ns	-	5ns	-

Filter Construction Performance

Operation	Prax (TypeLevel)	Prax (Runtime)	Diesel	Notes
Simple filter	2.1ns	7ns	4.7ns	DirectSql: 2.1ns
AND (2 filters)	4.3ns	17ns	5ns	DirectSql matches Diesel
AND (5 filters)	5.1ns	32ns	5ns	TypeLevel = Diesel!
AND (5) SQL gen	17ns	-	-	DirectSql write
IN (10 values)	3.8ns	21ns	14ns	Pre-computed pattern
IN (32 values)	5.0ns	-	-	Pre-computed pattern
IN (100 values)	158ns	160ns	-	Looped generation

Database Execution (PostgreSQL with Pooling)

Operation	Prax	SQLx	Diesel-Async	Winner
SELECT by ID	193µs	276µs	6.18ms*	Prax
SELECT filtered	192µs	269µs	7.40ms*	Prax
COUNT	255µs	320µs	-	Prax
SELECT prepared	191µs	-	-	Prax

* Diesel-Async establishes a new connection per iteration (~6ms overhead). Prax and SQLx use connection pooling with warmup.

Memory Footprint

Type	Size	Notes
Filter	64 bytes	Fits in single cache line
ValueList	24 bytes	91% reduction from SmallVec
FieldName	24 bytes	Cow<'static, str>

Optimization Techniques

DirectSql Trait (~5ns)

Zero-allocation SQL generation directly from typed filters.

Pre-computed Placeholders

256-entry static lookup table for PostgreSQL placeholders ($1, $2, ...). Pre-computed IN patterns for 1-32 elements.

Execution Plan Cache

Cache query plans with performance hints and automatic execution time tracking.

Zero-Copy Row Deserialization

Borrow string data directly from database rows without allocating.

Pipeline Execution

Combine multiple queries into a single database round-trip.

Type-Level Filters (And5, Or5)

Stack-allocated filter composition matching Diesel's zero-cost abstractions.

Why Prax is Fast

Cache-Friendly Layout

Filter enum fits in a single 64-byte cache line.

Zero Allocation Paths

Static field names and pre-computed values avoid heap allocation.

Connection Pool Warmup

Pre-establish connections and prepare statements at startup.

Prepared Statement Caching

All queries use prepare_cached() for per-connection statement reuse.

Prax vs Diesel

Where Prax Wins

✓ 6-7x faster SQL string construction
✓ TypeLevel filters match Diesel (5.1ns vs 5ns)
✓ 30% faster database execution vs SQLx
✓ Runtime flexibility with Prisma-like API
✓ Pre-computed IN patterns (3.8ns for 10 values)

Where They're Equal

≈ Type-level AND(5): 5.1ns vs 5ns
≈ DirectSql trait matches zero-cost abstractions
≈ 64-byte Filter vs Diesel's type-level size