reflow branch: cursor positions get set to -1,-1

[Björn Linse]

Libvterm gives the following assertion error

ASSERT fail: resize moved cursor outside the field to (-1,-1)

which causes subsequent errors, i e attempting to fetch the cell gives an invalid cell with width zero, which causes an infinite loop in pangoterm.

This happens after after cat:ing some text and then returning to zsh shell prompt, and then making the terminal very narrow (narrower than the shell prompt, roughly)

[Paul "LeoNerd" Evans (leonerd)]

> then making the terminal very narrow (narrower than the shell prompt, roughly)

I think this may be the killer part. It's about making the window narrower than the current cursor position. It doesn't normally break for me but my $PS1 ends "...\n$ " whereas if I set one that doesn't have a near-end linefeed, then I can get mine to break a lot earlier, in the same way.

[Jörg (joergrichter)]

I've got the same problem (state->pos.row is negative). This happens because state.cc:onresize calls screen.cc:resize (with callback) and passes on the VTermStateFields with the current cursor position.
This position is modified in resize_buffer to a negative value (statefields->pos.row -= (old_row + 1))
After the callback the fields are copied back to state->pos.

I fixed it with the following patch:

--- a/gui/vterm/state.c
+++ b/gui/vterm/state.c
@@ -1720,8 +1720,12 @@ static int on_resize(int rows, int cols, void *user)

   if(state->pos.row >= rows)
     state->pos.row = rows - 1;
+ else if(state->pos.row < 0)
+ state->pos.row = 0;
   if(state->pos.col >= cols)
     state->pos.col = cols - 1;
+ else if(state->pos.col < 0)
+ state->pos.col = 0;

   updatecursor(state, &oldpos, 1);

[Jörg (joergrichter)]

By bug was not with the reflow-branch. And it is already fixed in the trunk. Sorry for the noise