Follows #6592. Paste now also skips the names that effects in the
document's brush art use when it picks a free name for a pasted SVG
filter. When no free name keeps the filter within the 64 KB limit (a
one-letter name at the limit with every free letter and digit taken),
the paste is refused instead of keeping the name of the document's
filter. Clipboard::import_into and import_text_styles return a Result,
and their callers pass the error on before the document changes.
An exception that a finally block passes on, or one that leaves a
built-in's callback, no longer becomes $.error again over an error thrown
after it; throwing a caught error again still makes it the last one.
$.sleep with more than 0 inside a getter, a conversion or a built-in's
callback now throws whether or not the platform has a clock, so the web
build throws there as the other builds do.
Paste renamed a pasted SVG filter to a name no filter of the document
held, but an effect could still use that name, for example after its
filter was deleted. Defining a filter of that name later changed both
effects. Paste now also skips the names the document's effects use,
read only when a pasted filter needs a new name.
A new name with a tab, line feed or carriage return was written into
the filter's id as is, which XML reads as a space, so the id no longer
matched the effect's name. svg_filter::with_id now writes those
characters as character references, and paste takes a name only when
the filter's id reads back as that name.
`leading: "12abc"` set 12 pt leading and returned success, while `size`
and `tracking` in the same command refused trailing text. `leading` now
reads its value with `length_given`, so a string must be a whole number
or a length with a unit. The error message is unchanged.
The compiler now keeps a function declaration's name with its code, so
$.stack no longer shows [anonymous] for a declared function that is
called after the function that declared it has returned. The member test
expects "Web" for $.os in the wasm build, as the engine gives there.
The side limit compared the region's height in points times the scale
with 32,768. Far from the origin the snapped region can come back a
fraction of a pixel over, so art that the renderer draws at exactly
32,768 pixels was refused (a rectangle at y = 100,000 rasterized at
400 ppi). The check now uses the renderer's own rounded pixel width and
height.
Saving a legacy .ai of version 7 flattens each transparent image at its own
resolution, limited only by total pixels. The renderer draws at most 32,768
pixels a side, so an image that came out longer than that at its resolution
was cut off and placed as if whole: the part past that length showed the art
under it without the image. The resolution is now lowered until the longest
side fits.