Performance

Patitas is designed for consistent, predictable performance.

Why State Machine?

Regex-based parsers have unpredictable performance:

State machines guarantee O(n) regardless of input.

Benchmarks

Tested on MacBook Pro M3, Python 3.14t:

Optimization Strategies

1. Zero-Copy Text

The renderer extracts text directly from source using slices:

# Instead of copying strings
text = source[start:end]  # Zero-copy slice

2. StringBuilder

Output usesStringBuilderfor O(n) concatenation:

# O(n) total
builder = StringBuilder()
for part in parts:
    builder.append(part)
result = str(builder)  # Single allocation

3. Frozen Dataclasses with Slots

@dataclass(frozen=True, slots=True)
class Node:
    # 40% less memory than regular classes
    # Faster attribute access

4. Tuple Children

Using tuples instead of lists:

children: tuple[Inline, ...]  # Immutable, hashable

Memory Usage

Free-Threading Performance

With Python 3.14t (GIL disabled):

Near-linear scaling due to no shared mutable state.