Remove mechanism for catching proc_pidinfo and statfs timeouts. It was causing crashes, but I don't understand why. Perhaps mixing ARC and non-ARC code has some problem? I saw lots of crashes on 3.1.4 where proc_pidinfo exceeded the timeout and when it finally returned there was heap corruption. It looked as though the dispatch group had been freed even though there was an outstanding reference to it. At any rate, if proc_pidinfo is hanging you're going to have a bad time even without this workaround.

Add a beta build configuration, and use it for betas. Change beta, debug, and nightly build configurations to set the preprocessor BETA=1. Add ITBetaAssert, which crashes only in those build configurations. Add some ITBetaAsserts to find why we sometimes try to get negative line numbers.

Add iTermCallWithTimeout since 10.13 has a lot of system calls that never return. Prevent proc_pidinfo for hanging for more than half a second.

Refactor timer stuff into a single file.

Defer decreasing the frame rate until the next frame is drawn to try to avoid issues like 5928.

- Enable it for nightly build users
- 1% of lookups are checked for correctness, leaving a crashlog on failure
- Clear the DWC cache when dropping a line. I beleive this was the cause of the assertions in beta 6 and 7.
- Make the check for nightly build faster

searching all sessions with current selections for the one whose selection
was created most recently. This way even if the selection is emoved, you
can still paste it. Brought up in email, subject line "Keep selected text
in buffer until I select another text (like Mac terminal does)".

Prevent text extractor's line cache from outliving the vcall to performBlockWithLineCache, which caused a test to access freed memory

- Resizing a session would clear the find state, removing search results until the next search
- Clearing a buffer would remove highlights but not search results so they wouldn't reappear

Issue 5865.

Add a cache to LineBlock that remembers where double-width characters that get moved to the next line exist. The cache is off by default except for nightly build users. For others it can be enabled with an advanced pref. Issue 5855.

When you drag a tab from the tabbar into a split pane, that causes removeTab: to be called. It tries to create a restorable session to undo closure, but because no sessions actually get closed in this operation we create an empty shell of a restorable session. When it's restored, you get a zombie tab with no view. This commit ignores such operations and declines to push a restorable session on the stack. Issue 5785.

sequences. Add a python example to receive them and a shell script to send
them.

Rewrite the code for finding the current foreground job. It now builds a tree of processes and finds the foreground job with the deepest path from the root. It is possible for a foreground job to have a child foreground job. Issue 5730.

Also fix the menu opener to not select the last item.

Release URL codes when they're no longer needed. A mark is placed at the beginning and end of the URL. The one at the beginning is needed to keep the reference count to the code around until the one at the end is added. These marks are saved and restored.

OSC 8 ; url ST Anchor text OSC 8 ; ST

Don't code sign tests. This makes tests work locally. I wonder if it will break Travis? I can't overstate how much I hate this.

The first time the tmux gateway is auto-buried, open the sessions menu and point out how to restore it with a popover hanging off an invisible high-level window.

Get rid of cocoapods. Get rid of ocmockito. Get rid of ochamcrest. Now I understand the project again. Also, the tests now build.

Add a fake user defaults class to mock out the one thing that needs mocked.

It didn't work because sparkle uses its own version comparator to detect downgrade attacks.

This reverts commit 512c5b4a.