mirror of
https://github.com/mvschwarz/openrig-studio.git
synced 2026-11-07 05:56:21 -10:00
mvschwarz
9c824a7b06
studio-impl found this by building against it: the shell renders at most six app
actions and `decl.actions.slice(0, 6)` discarded the rest without a word. A surface
declaring seven rendered six and had no way to discover it — not in a DOM it cannot
see, not in any response, not anywhere. The seventh button simply never worked,
which reads as a bug in the app rather than a limit of the shell.
The bound itself is right and stays: the left half shares one fixed bar with the
title, the crumb and the shell's own half, so an unlimited list pushes into
controls the surface does not own. The silence was the defect.
Told on BOTH channels, because they reach different people. A console warning
naming the dropped ids reaches whoever is building the app right now. A
`{t:"header-overflow", limit, declared, dropped}` message reaches THE APP, which is
the only party that can adapt — and "an app cannot see the cap to respect it" was
the actual complaint, which a doc alone would not have answered.
The number is now a named constant rather than a literal inside a slice. That is
why it went undocumented: it existed in exactly one place with no name to look up
and nothing for a doc to point at. A committed test reads the shipped limit and the
documented one and compares them, so the two cannot drift.
WHILE DOCUMENTING IT I FALSIFIED A NEIGHBOURING SENTENCE AND FIXED IT:
shell-protocol.md said `markup` "is the one message that travels in that
direction". Adding a second made that false in the same commit that made it false.
Verified in a live shell rather than reasoned about — `tellSurface` is a const
declared after renderAppHeader's first caller, and this file has shipped four
TDZs. A surface declaring eight actions rendered six, received
{limit:6, declared:8, dropped:["a7","a8"]}, and the console carried the warning
naming them. The committed checks are structural and say so, each with a control.
290 tests.