get_update_time was returning -1 in some cases even if schedule_update
was called prior. So the master clock was overshooting the correct update
time and missing some frames. This seems to be happening most under moderate
CPU load (gnome-shell#1411 (closed)), but not so high a load as to justify the frame
skips that were occurring.
The reason for get_update_time returning -1 in these cases was because
it was considered impossible to calculate the next update time while
pending_swaps > 0. That's not completely true. We know if
pending_swaps == 1 then schedule_update should schedule one frame later.
So just factor pending_swaps into schedule_update. Now get_update_time
won't return -1 in these cases so we're more likely to render the next
frame (or move the cursor) at the correct time.
Closes: gnome-shell#1411 (closed)
Prerequisite (just FYI for backporters): !216 (merged)