modal + file_pane: rework editing math and pane behavior, config/syntax additions, unit tests

2 files changed, 6 insertions, 7 deletions

diff --git a/test/snapshots/badutf.golden b/test/snapshots/badutf.golden
index dceef522..fda9f8a7 100644
--- a/test/snapshots/badutf.golden
+++ b/test/snapshots/badutf.golden
@@ -32,7 +32,7 @@
 == snap hscroll-midglyph grid=50x30 cursor=46,2
 |New Newcol Find Grep Help Tutor Dump NextColor Deb
 |  /tmp/pardes-snap/badutf/cwd/wide.txt Save New De
-|     1 ��語 日本語 日本語 日本語 tail
+|     1  日本語 日本語 日本語 日本語 日本語 tail
 |     2
 |     3
 |
diff --git a/test/snapshots/badutf.snap b/test/snapshots/badutf.snap
index 49d90bdd..46592178 100644
--- a/test/snapshots/badutf.snap
+++ b/test/snapshots/badutf.snap
@@ -1,7 +1,7 @@
 # Invalid UTF-8 must RENDER, never panic. Two sources, both ordinary use:
 # a file pane holds whatever bytes are on disk (latin-1, a truncated sequence,
-# an ELF opened by mistake), and the byte-column hscroll cuts a multi-byte
-# glyph in half by design. Undecodable bytes come out as U+FFFD, one per byte.
+# an ELF opened by mistake). Undecodable bytes come out as U+FFFD, while
+# horizontal scrolling must start at a complete grapheme boundary.
 file bad.txt hello \xff\xfe world\nlatin-1 caf\xe9 tail\ntruncated \xe6\x97 here\nend\n
 file wide.txt \xe6\x97\xa5\xe6\x9c\xac\xe8\xaa\x9e \xe6\x97\xa5\xe6\x9c\xac\xe8\xaa\x9e \xe6\x97\xa5\xe6\x9c\xac\xe8\xaa\x9e \xe6\x97\xa5\xe6\x9c\xac\xe8\xaa\x9e \xe6\x97\xa5\xe6\x9c\xac\xe8\xaa\x9e \xe6\x97\xa5\xe6\x9c\xac\xe8\xaa\x9e \xe6\x97\xa5\xe6\x9c\xac\xe8\xaa\x9e tail\nsecond line\n
 start 30 100 bad.txt
@@ -9,15 +9,14 @@ wait 8000 New Newcol
 stable 700 20000
 snap badbytes
 # 50 columns, not 100: a FILE boots alone and fills the window, and at 100 the
-# whole line fits, so `$` scrolls nowhere and there is no glyph to cut. This is
+# whole line fits, so `$` scrolls nowhere. This is
 # the width the left column used to be, i.e. the pane this case was written for
 start 30 50 wide.txt
 wait 8000 New Newcol
 stable 700 20000
 # SPC t w turns soft wrap OFF: hscroll is the other half of this case and a
-# wrapped body has nothing to scroll sideways, so the byte cut that halves a
-# glyph only exists here. (The WRAP cut halves one too — same bytes, same
-# U+FFFD — and wrap.snap pins that one.)
+# wrapped body has nothing to scroll sideways. The unwrapped view must keep the
+# first visible CJK glyph intact instead of starting inside its UTF-8 bytes.
 key space t w
 stable 400 5000
 key $