Previously we used the system’s temporary directory and removed the script after use.
This is because macOS 10.15 will “call home” (notarization?) when executing a new script (that it has not seen before), so with an unstable network connection, running commands could take several seconds.
Apparently the window’s delegate is not used for context menu actions in full screen mode, although NSApp’s targetForAction: does find the receiver, so this appears to be a framework bug.
The specific issue being addressed is macOS 10.15 where the mount point for ~/Library/Application Support is under /System/Volumes/Data so our assertion that the relative path received in our callback can be appended to the mount point, to match the path being observed, is false.
As a workaround we call realpath() on the path we wish to observe, observe that path, and in our callback we calculate the relative path from the real observed path, and append that to the requested path.
In theory the caller could be made responsible for expanding paths to their real path, but the current abstraction allows observing non-existing paths, which is why it needs to handle this path rewriting.
If one event cycle triggered multiple re-layouts, we could end up reusing a tab bar view that had been removed from its superview because the weak pointer had not yet been cleared (as the view hadn’t yet been fully released).
This could happen during window construction in a scenario like this:
1. Window gets created at some size, tab bar views are created
2. Window size gets restored causing some tab bar views to be removed
3. Active tab gets restored bringing an overflow tab into the visible area, re-using a removed view
We don’t yet rely on any 10.12 specific API other than os_log which can be skipped without user impact.
Unfortunately when trying to build for 10.10 using latest Xcode/SDK the linker gives an error for missing symbol: _objc_loadClassref.
Going forward we will require 10.12 but as there is no non-beta build for 10.11 users yet, I wanted to make one.
This is because we rely on `-X` (skip extended attributes / resource forks) which is only available with Apple’s version of `cp`, and it is not unlikely that the user has GNU’s version of `cp` available via PATH.
The issue appears to be that accessing the clipboard, as part of setting up a normal document window, will stall when trying to connect to the system’s clipboard server.
Here’s the relevant stack trace:
-[OakPasteboard checkForExternalPasteboardChanges]
-[NSPasteboard stringForType:]
-[NSPasteboard _dataForType:securityScoped:]
-[NSPasteboard _dataForType:index:usesPboardTypes:combinesItems:securityScoped:]
-[NSPasteboard _dataWithoutConversionForTypeIdentifier:securityScoped:]
CFPasteboardCopyData (in CoreFoundation)
___onqueue_CFPasteboardRequestDataFromDaemon_block_invoke.341
usleep (in libsystem_c.dylib)
nanosleep (in libsystem_c.dylib)
__semwait_signal (in libsystem_kernel.dylib)
For example if we have a directory with files named 001-foo.txt, 002-bar.txt, and 003-fud.txt, duplicating one of these will result in the duplicate starting with 004.
Also improve support for finding dates in file names being duplicated. Previously we only looked for YYYY-MM-DD but now we also support YYYY_MM_DD and YYYYMMDD. If a date is found in the file name being duplicated, it is replaced with the current date in the duplicate.
The issue is that the types requested may be ‘NSStringPboardType’ but that symbol is deprecated (starting with 10.14), yet the replacement (‘NSPasteboardTypeString’) is not string-equivalent, therefore our simple string comparison would fail.
This effectively broke services starting with 1ef78b3af7 (where we adopted the new symbols).
Implementation of validRequestorForSendType:returnType: does not appear to be a problem, as that method is called with both ‘NSStringPboardType’ and ‘NSPasteboardTypeString’.
Also switched to using NSPasteboard’s writeObjects: so we don’t have to explicitly deal with the type of the data we write to the pasteboard (presumably NSString will declare it using both the old and new type).
This is to avoid redundancy as ninja_required_version is hardcoded in the generator script and builddir is already passed via the --build-directory/-C option (and explicitly exported as a variable in the generator script).
There may be situations in which the user already have (generated) ninja build files which do not match our assertions.
This reverts commit b42cbb1f7e.
