Séquences paresseuses et flux infinis

A lazy sequence computes its elements only as they are needed. This lets you describe potentially infinite data and consume just the part you want.

range with no arguments returns an infinite lazy sequence. It never materializes fully; you take only what you need.

iterate builds an infinite sequence by repeatedly applying a function to a starting value.

repeat yields the same value forever; cycle loops a collection forever. Both are lazy.

You build your own lazy sequences with lazy-seq. The body runs only when the next element is requested.

map, filter, and take-while are all lazy. Chaining them does no work until you consume the result.

Lazy seqs are realized in chunks of 32 for efficiency. So asking for one element may compute up to 32. Keep transformations side-effect-free to avoid surprises.

Use doall to fully realize a lazy sequence (e.g. to trigger side effects), or dorun when you do not need the results.

If you keep a reference to the head of an infinite seq while walking it, the whole realized portion stays in memory. Let go of the head to allow garbage collection.

Lazy sequences elegantly express recursive streams like Fibonacci numbers.

Lazy seqs build intermediate sequences; transducers (from a prior lesson) avoid them. Use laziness for infinite/streamed data and transducers for high-throughput pipelines.

Test your lazy-sequence intuition.

You learned how lazy sequences defer computation and model infinite streams.

• Generate with range, iterate, repeat, cycle, lazy-seq
• map/filter stay lazy until consumed
• Beware chunking and head-holding