- A knockout, clip or blend-mode merge being tried may use the whole
import allowance beyond what was read before it. A merge that is made
must still fit the allowance with everything read before it.
- A knockout that turns out to cut nothing moves what its drawing read
out of the allowance, as a merge that isn't made does. A hidden one
also forgets what its drawing counted, so a later merge that leaves it
out reports only the hidden layer.
- A hidden layer in a mode VectorCraft lacks is left out before any
knockout drawing, as before the knockout merge.
- A knockout at 0 fill whose painted pixels lie outside the bounds of
every layer below it is kept apart before their canvas is made.
- A knockout at 0 fill whose pixels cover none of the painted pixels below
it is no longer merged with them.
- A hidden knockout is left out only when its merge would be made or is
too large, so one whose layer mask hides all of it comes in with
hiddenLayers. What drawing it counted is forgotten when it is left out.
- The pixels and canvases of a merge that isn't made are counted against
a separate limit of the same size, so the layers can still be read
apart afterward.
- docs: the memory-limit fallback for knockouts is listed with the
knockouts that aren't applied.
`transform` read `reflect` with `as_str`, `copy` with `as_bool` and
passed `origin` through unchecked. A `reflect` of `42` or `true`
skipped the reflect step and reported success for the steps before it,
a `copy` of `"yes"` transformed the original instead of a copy, and an
`origin` that was not two numbers fell back to the selection centre.
The handler now checks all three before it selects `ids` or runs a
step, and returns an error that names the argument. A null value of
any of the three is left out. The tool description and the
`transform` row in docs/mcp.md say so.
Fixes#6448
A point counts as on the path's outline when it is less than the
touching tolerance from it, with its distance from the path's center in
place of the distance between centers. The params text, docs/mcp.md and
the progress doc now say so, and say that every Tangents choice in the
dialog gives the same lines for a point.
Tests: a point half the tolerance inside and outside a circle gets the
one line there, and farther off gets none or two; a point placed on a
turned outline gets a line as long as the outline is wide along it; the
dialog previews the same two lines for every Tangents choice and keeps
them as one step, and a point inside changes nothing and says why.