apply-flyweight-pattern

Apply Flyweight Pattern

Share immutable intrinsic state across many fine-grained objects to reduce memory usage when the number of object instances is very large.

Why This Is Best Practice

Adopted by: Java's String pool (String.intern() — every JVM uses it for string literals, reducing string object count by 60–80% in typical applications), Python's integer interning (integers -5 to 256 are singletons in CPython), Java's Character cache (characters 0–127), and every major game engine's particle system (10,000+ particles sharing a single mesh/texture object). Impact: GoF documents a text editor example where storing one glyph object per character in a 100,000-character document would require 100,000 objects. With Flyweight, one object per distinct glyph character (typically ~100 unique glyphs) serves all 100,000 positions — a 1000× reduction in object count. Why best: The alternative — one full object per instance — is correct and simple but prohibitive at scale. Flyweight applies only when profiling confirms object count is the memory bottleneck; it trades simplicity for a controlled reduction in allocation cost.

Sources: Gamma et al. (1994) pp. 195–206; Java String pool documentation; CPython source (Objects/longobject.c integer cache)

Steps

Step 1: Separate intrinsic (shared) from extrinsic (context-specific) state

Intrinsic state (shared): character glyph image, font, size
Extrinsic state (per-instance): position on page (x, y), color

Particle system:
Intrinsic: mesh, texture, animation frames
Extrinsic: position, velocity, lifetime

Intrinsic state must be immutable — it's shared across all instances that use it.

Step 2: Make the flyweight class hold only intrinsic state

class GlyphFlyweight:
    def __init__(self, character: str, font: str, size: int):
        self.character = character   # intrinsic — shared
        self.font = font
        self.size = size

    def render(self, x: int, y: int, color: str) -> None:
        # x, y, color are extrinsic — passed in, not stored
        print(f"Render '{self.character}' at ({x},{y}) in {color} [{self.font} {self.size}pt]")

Step 3: Build a factory that returns shared flyweight instances

class GlyphFactory:
    _pool: dict[tuple, GlyphFlyweight] = {}

    @classmethod
    def get(cls, character: str, font: str, size: int) -> GlyphFlyweight:
        key = (character, font, size)
        if key not in cls._pool:
            cls._pool[key] = GlyphFlyweight(character, font, size)
        return cls._pool[key]

Step 4: Store extrinsic state outside the flyweight — in the client context

class TextDocument:
    def __init__(self):
        self._chars: list[tuple[GlyphFlyweight, int, int, str]] = []

    def add_character(self, ch: str, font: str, size: int, x: int, y: int, color: str):
        glyph = GlyphFactory.get(ch, font, size)   # shared flyweight
        self._chars.append((glyph, x, y, color))   # extrinsic state stored here

    def render(self):
        for glyph, x, y, color in self._chars:
            glyph.render(x, y, color)

Step 5: Measure first — only apply where profiling shows object count is the bottleneck

import sys
# Before: 100,000 GlyphFlyweight objects
# After:  ~95 unique glyphs in pool
print(f"Pool size: {len(GlyphFactory._pool)}")
print(f"Pool memory: {sum(sys.getsizeof(g) for g in GlyphFactory._pool.values())} bytes")

When NOT to Use

• When object count is small — if an application creates hundreds, not tens of thousands of objects, the pattern adds complexity with no measurable benefit.
• When extrinsic state is large — if the per-instance state dominates memory usage, sharing the intrinsic portion saves little. Profile before applying.
• When objects must maintain identity — if callers compare objects by identity (is) rather than value, sharing breaks expected behavior.

Common Mistakes

Storing mutable state in the flyweight. If the "shared" state changes per use, it is extrinsic, not intrinsic. Mutable flyweight state causes one caller's changes to silently affect all others sharing the same object.

Not using a factory. Flyweights only save memory if they're actually shared. Without a factory/pool, each caller creates a new instance and the pattern achieves nothing.

Applying prematurely. Flyweight is a memory optimization, not an architectural pattern. Apply it after profiling confirms object count is the problem, not before.