macos: the AppKit shell, its icon, and the offscreen e2e harness

1 files changed, 63 insertions, 0 deletions

diff --git a/test/macos-snapshots/rotate.snap b/test/macos-snapshots/rotate.snap
new file mode 100644
index 00000000..17b239f4
--- /dev/null
+++ b/test/macos-snapshots/rotate.snap
@@ -0,0 +1,63 @@
+# pardes_rotate spends a two-finger twist as the search-step keys.
+#
+# AppKit reports rotation counterclockwise-positive and the core reads clockwise
+# as `n`, so a NEGATIVE delta is the FORWARD step. That inversion is exactly the
+# kind of thing a unit test on the accumulator cannot catch and a golden can:
+# what the screen does is the assertion.
+#
+# One notch is 20 degrees (rotation_notch_degrees in src/macos.zig) and the
+# remainder is banked, which makes the pair below exact rather than approximate:
+# -25 spends one notch and banks -5, then +25 lands on 20 and spends one back.
+start 30 100
+wait 8000 New Newcol
+wait 8000 $
+stable 700 20000
+text printf 'x MA''RK a\nzz\nx MA''RK b\nzz\nx MA''RK c\n'
+key enter
+wait 10000 MARK c
+stable 700 10000
+key c-b
+stable 600 8000
+# `/` types the pattern into the tag; Enter searches the pane's scrollback and
+# writes the three hits into a +Search buffer, which is what n/N walk.
+text /
+stable 400 5000
+text MARK
+stable 400 5000
+key enter
+wait 10000 @p0:
+stable 700 15000
+snap results
+# A gesture beginning re-zeros the dial, so travel left over from an earlier
+# twist cannot make the first degree of this one jump a match.
+rotate 0
+# Clockwise: one notch of `n`, forward onto the first hit and selecting it.
+rotate -25
+stable 700 15000
+snap forward
+# ...and again, onto the second. Two steps out rather than one, because the
+# first hit is where the list already starts: stepping BACK from it has nowhere
+# to go, and a golden that cannot move is a golden that proves nothing about
+# which way the dial turns.
+rotate -25
+stable 700 15000
+snap forward2
+# A new gesture, because the dial banks its remainder exactly like the scroll
+# accumulator does: two 25-degree notches leave -10 behind, so reversing INSIDE
+# the same twist would have to spend that first and 25 back would move nothing.
+# That hysteresis is wanted — it is what stops a thumb resettling on a notch
+# boundary from flip-flopping between two matches — and lifting the fingers is
+# how a hand clears it.
+rotate 0
+# Counterclockwise: `N`. This must land back exactly on `forward`.
+rotate 25
+stable 700 15000
+snap back
+# And the other half of the contract: a twist IS the keystroke, so typing the
+# key the dial claims to send must go the same way the dial went. `n` here must
+# reproduce `forward2` — if the dial were sending something else, or the sign
+# were inverted, these two would part company.
+text n
+stable 700 15000
+snap forward_by_key
+draw rotate