-- mset accumulator: `m = mset m k v` in a loop runs in O(n) amortised, not O(n²).
-- The compiler peephole rewrites the self-rebind to an in-place HashMap update
-- when the variable holds the sole reference, matching the `s = +s c` and
-- `xs = +=xs v` accumulator patterns.
-- Build a word-frequency map by accumulating into the same variable. The
-- pattern below used to clone the entire map on every `mset` — fine for a
-- handful of keys, OOM at corpus scale (nlp-engineer's 222k-token Moby Dick).
count-words ws:L t>n;m=mmap;@w ws{c=mget m w ?? 0;m=mset m w +c 1};len (mkeys m)
-- Three-word demo: ["the", "cat", "the"] gives 2 distinct keys, "the" mapped
-- to 2 and "cat" mapped to 1.
demo>n;count-words ["the","cat","the"]
-- Distinct-key accumulation: 200 keys via fmt.
build-200>n;m=mmap;@i 0..200{k=fmt "k{}" i;m=mset m k i};len (mkeys m)
-- run: demo
-- out: 2
-- run: build-200
-- out: 200