summaryrefslogtreecommitdiff
path: root/docs/next-winmode.md
diff options
context:
space:
mode:
authorGabriel Schneider <[email protected]>2026-10-24 13:48:41 -0300
committerGabriel Schneider <[email protected]>2026-10-24 15:41:50 -0300
commit32fd7e0cd8b472573cc9e2d49934245fb279d412 (patch)
tree9aefab38150184e4c64aa5b8dc4ed13461617f85 /docs/next-winmode.md
parent578710b72755503ef2ef40a0ae301e608af3e3af (diff)
downloadpardes-32fd7e0cd8b472573cc9e2d49934245fb279d412.tar.gz
pardes-32fd7e0cd8b472573cc9e2d49934245fb279d412.zip
Docs for window mode's d and c and the delayed Del hint: Help, cheatsheet, guide, reference, tutor, helix-keys
Diffstat (limited to 'docs/next-winmode.md')
-rw-r--r--docs/next-winmode.md69
1 files changed, 69 insertions, 0 deletions
diff --git a/docs/next-winmode.md b/docs/next-winmode.md
new file mode 100644
index 00000000..c2a01da8
--- /dev/null
+++ b/docs/next-winmode.md
@@ -0,0 +1,69 @@
+# winmode: d and c in window mode, the Del question as a hint
+
+> The window mode should have d for deleting and c for collapsing too, the
+> deletion should keep that j/k movement. The current spc d j/k should only
+> show the message after a timeout so that a user that does this fast won't
+> even see it, it will be more like a hint for users that tried to delete
+> and nothing happenend.
+
+## As built
+
+- `layout.windowKey`: `c` runs `Collapse` on the outlined pane; `d` makes
+ the outlined pane active and runs `Del` on it, exactly the keyboard Del
+ (`can_ask` is set: it is a key). With panes above and below, Del asks
+ and the next key is its answer (`Pardes.answerQuestion`, which already
+ runs before window mode sees keys); with one neighbour it closes at once.
+ Either way, once the pane is gone the outline moves to `p.active`, which
+ `removePane` already sets to the pane that grew into the rows. The mode
+ stays on: `Ctrl-w d j d` closes two panes. A refused Del (unsaved text)
+ leaves the outline where it is, the refusal on the message row.
+- Keys: `config.window_delete` (`d`) and `config.window_collapse` (`c`),
+ apart from `window_keys`, because those are also the header walk under
+ `Ctrl-w` on a tag, where `d`/`c` mean nothing.
+- Labels: `layout.window_labels` loses `d` and `c` (22 -> 20 letters).
+- The hint: Del sets `Pane.question_due_ns = clock + 400 ms` when it asks;
+ `Messages.collectNotices` leaves the prompt band out while
+ `question_due_ns > now_ns`; `nextWake` wakes once at that time (while
+ `question_due_ns > stepped_ns`, so stepping past it ends the wake and a
+ tick draws the frame). Same pattern as the REPL's `repl_enter_due_ns`.
+ The question exists from the first moment: keys answer it, the 9P log
+ says `ask` at once, a script's `answer` works; only the band waits.
+
+## Why 400 ms
+
+Typed in one go, the `j` after `SPC d` lands 100-250 ms after the `d`
+(a leader sequence's inter-key gap). Above ~300 ms it is a pause, the hand
+waiting to see something happen. 400 ms clears a relaxed typist with margin
+and is still quick enough to read as "it is asking", not "it is broken".
+Key-hint popups elsewhere (which-key) wait a few hundred ms for the same reason. It is a compile-time
+constant beside `del_question` (keys are compiled in); a Runtime setting
+would be one line in config if the author wants to tune it live.
+
+## Tests
+
+- `window mode d deletes the outlined pane and asks k or j, c collapses
+ it, and the mode stays` (pardes.zig): c twice restores the height; d asks,
+ Esc keeps it, d j gives the rows below and the outline follows; d with one
+ neighbour closes at once onto it; labels hold none of h j k l d c.
+- `the Del question shows only once its delay is up: a quick SPC d j never
+ sees it`: no prompt band before the due time, nextWake at it and not
+ past one frame, the band after, no wake once stepped past; a fast
+ `SPC d j` closes with the band never showing it.
+- The old Del test moves the clock to the due time before looking at the
+ band. `test/snapshots/delask.snap` waits out the delay (`settle 700`)
+ before `snap asked`; its golden is unchanged (it raced otherwise: the
+ first `stable 400` saw no change and captured the band empty).
+
+## Open questions, positions
+
+- Should `d` leave window mode after closing? Position: no. The point of a
+ mode is a run of layout edits (`Ctrl-w d j d c`), and the outline landing
+ on the pane that grew says where you are; Enter/Esc end it as ever.
+- `d`/`c` taken from the labels: the 21st and 22nd panes lose a letter.
+ Position: fine; past 20 panes you move anyway. The alternative (labels
+ only after a prefix) costs a key on every jump.
+- Del's question from a click or 9P never asks, so the delay touches only
+ the keyboard paths that ask today.
+- Pushback: the hint could be the message row's own lingering text instead
+ of the prompt band, but the band is where the answer's keys are named and
+ the caret logic already knows it; kept.