[hlcfan]

Fix #11462

Bug

When user clicks on a purely prompt line without input area, it moves the cursor
to the start of the input. For example, I use starship prompt, it shows a whole
line for the prompt, e.g.

{username} {cwd} // prompt line
> ls -lah        // input

When user double clicks on the prompt area to copy text, the cursor shouldn't
move.

Fix

In promptClickLine, when user clicks on a prompt area, it traverses down the
prompt rows. If one of the rows contains input, allows the cursor movement.
Otherwise, if pure prompt line, return zero movement.

AI Usage

I used Antigravity to pinpoint the prompt area clicking logic.

[mitchellh]

There's some stylistic Zig stuff here I want to fix but willing to do later.

[hlcfan]

Hi @mitchellh is this PR good to go or pending anything?

[PRIHLOP]

Hello.
I re-tested this PR against current main
(683d8db643b95cf229bfb5fe9fab9ae677920343).

Refresh required

The fix is still valid and the full promptClickMove test filter
passes after a small API refresh. Git merges the branch without
textual conflicts, but the result does not compile as-is because
three APIs changed since March.

The required API refresh is limited to these changes in
src/terminal/Screen.zig:

-const cells = row_pin.node.data.getCells(rac.row);
+const cells = row_pin.node.page().getCells(rac.row);

And in both newly added tests:

 const testing = std.testing;
 const alloc = testing.allocator;
+const io = testing.io;

-var s = try init(alloc, .{ .cols = ..., .rows = 5, .max_scrollback = 0 });
+var s = try init(io, alloc, .{ .cols = ..., .rows = 5, .max_scrollback_bytes = 0 });

Verification

I created a local merge of the PR into current main, applied only the three mechanical API adaptations above, and ran:

zig fmt --check src/terminal/Screen.zig
zig build test -j1 -Dtest-filter="promptClickMove"

Both completed successfully (exit 0). This filter includes the two PR regression scenarios and the existing promptClickMove tests:

• clicking a pure prompt-only row returns zero movement;
• clicking a wrapped prompt that contains input still moves to the input correctly;
• existing left/right prompt-click behavior remains green.

Environment:

• Linux 6.12.100+deb13-amd64, x86_64
• Debian 13 (trixie) container
• Zig 0.16.0
• GTK 4.18.6
• libadwaita 1.7.6

AI assistance disclosure: I used Hermes Agent(with LLM GPT 5.6 Sol) to inspect the current upstream state, prepare the local refresh, run the test matrix, and draft this report. I reviewed the changes and results.

Guys, could you please help with this PR @mitchellh @hlcfan

[hlcfan]

Thanks @PRIHLOP for reporting this. Updated the code according to the API change.

[PRIHLOP]

Hello @mitchellh
Could you please approve workflows to complete checks of these changes?

[PRIHLOP]

Hello @jcollie
Could you please help with review and approve (second is needed) of these changes?

[jcollie]

Should also add a test that has a starship-style prompt that looks like the prompt in the issue.

[jcollie]

@hlcfan can you rebase instead of merging main?

[hlcfan]

@jcollie Done.

[jcollie]

Approved pending CI

[hlcfan]

@jcollie @mitchellh is it good to go?