Writing an open file's bytes empties the file first. When the write then
failed, for example on a full disk, the file had lost its old bytes but
was left out of the files the run names as written. It is now named as
soon as it has been opened for the write.
`writeln` uses CRLF by default on Windows, and the script reports paths
without the `\?\` prefix that canonicalize adds there. The test expected
LF and compared with the prefixed path, so it failed on Windows. It now
expects the platform's default line end and compares plain paths.
`rename` matched open File objects by the path as given, while the objects
hold the names the file system keeps. Where case does not matter, a folder
named in another case was renamed but its open files kept the old path, so
their changes could not be written and they stayed under the old path's
rules for reading. The objects are now matched by the kept name, and a
rename whose kept name cannot be learned is refused.
Adds Danish rows for the script File and Folder messages, now that Danish is a complete-message language, and removes the two Danish rows for messages the scripts no longer give.
A file opened with open('r') kept its old bytes after copy wrote over it
or length changed it, while an object open for writing took the new
bytes. Both now update every object that holds the file. An object
that holds a renamed file is found at the new path, and an object whose
file was removed is not taken for a file made later at its path.
Taking the lock on Script Grants.lock blocked until the holder let go,
so another process that stalled while holding it stopped listing,
loading, keeping and revoking grants for good. The lock is now tried 20
times, 50 ms apart, and then fails with TimedOut: a change is not made
and a read gives no grants rather than reading without the lock.
`rename` moved the write path of open File objects but left the path they
were read from at the old name. When new code then withdrew the kept grant
for the folder, a grant the run still held for the old path let the object
go on giving the file's bytes. The read path now moves with the write path.
Keeping a second grant with Always marked all the code the run had
evaluated as recorded, so the run's end skipped the grants kept before
it, and they kept the code list from when they were kept. The next run
that evaluated that code withdrew them. Keeping a grant now writes the
current code list into the grants the run kept before it as well.
The four panel tests that export into a folder made it under the system
temporary folder and removed it at the end, so a failing assertion left
the files behind. They now use vectorcraft_testkit::TempDir, which
removes the folder when the test ends or panics.
The WebP chunk walk added the chunk length to the offset without a
checked add, so a length near u32::MAX overflowed on 32-bit targets
(wasm32) with overflow checks on. It now reads the chunk through the
same checked slice helper as the PNG, JPEG and TIFF readers.
Create Sub-folders and Open Location after Export are check items that set
the settings the panel shares with Export for Screens, and the panel's
Export follows them. Create Sub-folders is checked as the dialog opens on
it: on for a document without export settings, and on for remembered
folders per row, format or item, which the panel's Export now uses too.
Format Settings... opens Export for Screens on the Assets tab showing the
first row's format settings.
assets.settings.set takes openLocation (kept in document.exportSettings,
not in presets), refuses a subfolders or openLocation value that is not
true, false or null, and with subfolders false given alone also ends a
remembered folder per format or item. Export for Screens reads remembered
empty folders as none, as an export does.
This reverts commit bada8e1710. #7638 merged with its gate failing: besides six unformatted
files, render and trace don't compile against craft-libs e2dc681 (u64::from(usize) in
render/src/paint.rs, a private craft-libs function, and a call with one argument too many in
trace/src/opaque.rs), so main's CI fails. The tree is that of 0adbdce6cb, whose CI passed.
When a Type 2 CIDFont or a Type 42 descendant of a composite font has no FontMatrix, the
font a cshow procedure sees as current now takes the identity as its default, as glyph
measuring and drawing do, instead of the 1/1000 default of other fonts.