[neu-rah]

Three small data components, in the order of the commits. 0.1.19 -> 0.1.22.

OnChange<fn> (moved from OneMenu)

fn(v) on every set(), v re-read via get() after Base::set(). Unchanged behaviour: OneMenu keeps using oneData::OnChange; (follow-up PR in OneMenu) so its code and examples build unchanged. It is a data component, not a menu one.

OnSync<fn> (new)

Composed over a change tracker (Watch, Dirty): on sync() it calls fn(value) when changed() is true, then Base::sync(), so the tracker keeps its copy and is unchanged. For values that move outside set() (an ISR, a hardware register, a sensor read). Needs get(), changed() and sync() of what it wraps; no Type (so it composes over an ItemRef to a node).

DataDef<OnSync<publish>, Watch<Int>> d;   // d.sync() calls publish(v) once per change, not on an equal set

ReadOnly public, set() deleted

ReadOnly inherited its wrapped part privately, which hides the whole API of the base items, including what an ItemDef reaches through its chain (using Base::enabled), so a ReadOnly could not be composed into an item. set() is now deleted; get(), print(), changed() and sync() stay.

Tests (test/test.cpp)

OnChange (every set, also an equal one; chains through a Watch), OnSync (once per change, not at set(), not on an equal set, two in a row each fire once, over Dirty: it starts dirty and an equal set() counts), ReadOnly (set() not callable, changed()/sync() kept; the existing ReadOnly tests unchanged). g++ and clang++ with -Wall -Wextra -Wpedantic -Werror, and under ASan+UBSan.

Cost (AVR, atmega328p -Os, benchmark/avr_onsync.sh, simavr; eight sets)

	flash	RAM	sizeof	cycles
bare Data (set, read back)	592	46	1	107
Watch, loop tests changed() and syncs by hand	622	48	2	159
Watch + OnSync	622	48	2	159
Data + OnChange	602	46	1	147

OnSync costs nothing over the hand-written loop (the two images are byte-identical in size and cycles). Headers only: nothing is instantiated unless used.

🤖 Generated with Claude Code