Use AMIndeterminateProgressIndicator in tabs with the light theme instead of NSProgressIndicator because it's faster. See issue 4582.

Fix a bug where we returned NO when parsing a wrapped DCS code, which prevented further parsing from the buffer until another read happens. Issue 4588 (the part where you have to bang on the keyboard for a while after running the solarized script)

Fix a bug in -[PTYTask pidIsChild] that caused it to always return NO, which prevented child processes from getting waited on when not using session restoration. Issue 4612.

Fix a bug in download.sh where it wouldn't accept multiple files as command line arguments. Also print an error on directory files.

Fix a bug where we didn't store the previously active app when showing an already-open sticky hotkey window. Issue 4606.

Fix how we return the selected text range when there's no selection. It should be a zero-length selection at the cursor location. Add some debug logging for accessibility calls.

Fix a frequent crasher in 2.9.20160426 where moving a session to a split pane in the same tab caused it to get released prematurely, causing a crash.

Insert a .1 second delay between screen size changes when jiggleTTYSizeOnClearBuffer is enabled to fix issue 4494.

Change how automatic profile switching paths are globbed. Now /foo/bar/* will match a path of /foo/bar, where before it required you to be in a subdirectory of /foo/bar. The old rule can be rewritten as /foo/bar/*/* if that's what you really want.

For automatic profile switching: if the current profile outscores the topmost match on the stack, do not switch. Fixes issue 4581.

Add support for adaptive frame rates. Try to estimate the throughput. Use 60fps if it's under 1000 bytes per second, 30fps if over. Use the slower framerate when idle. It looks ugly when we're flip-flopping between frame rates but I don't know if that will be a problem in practice. This is off by default and behind an advanced setting.

Don't dismiss and re-show the captured output announcement. When there's a long compile with a lot of warnings it comes and goes which is annoying

Fix testReferenceRemovedFromNotificationCenter to actually test what needs to be tested, even when run on 10.11.

notification center on dealloc. Do not leak _class. If the weak reference's object gets its class
changed at runtime, respect that. I believe the crash would occur when a weak reference gets
released before its object. When the object is released, it'll post a notification that is received
by a now-dealloc'ed weak reference.

this didn't always crash. Maybe it always dies on older OSes since 10.11 automatically removes the
observer.