You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
patch 9.2.0253: various issues with wrong b_nwindows after closing buffers
Problem: close_buffer() callers incorrectly handle b_nwindows,
especially after nasty autocmds, allowing it to go
out-of-sync. May lead to buffers that can't be unloaded, or
buffers that are prematurely freed whilst displayed.
Solution: Modify close_buffer() and review its callers; let them
decrement b_nwindows if it didn't unload the buffer. Remove
some now unneeded workarounds like 8.2.2354, 9.1.0143,
9.1.0764, which didn't always work (Sean Dewar)
close_buffer() now doesn't decrement b_nwindows when not unloading buf, or when
buf isn't w_buffer after autocmds (they would've already decremented it).
Callers are now expected to decrement b_nwindows if w_buffer is not NULL after
close_buffer(), and when still intending to switch buffers or close win, for two
reasons:
- close_buffer() autocmds may have switched buffers. The new w_buffer's
b_nwindows would also need decrementing.
- After close_buffer(), callers may opt to not switch w_buffer or close win.
b_nwindows would need to be incremented again. (unless w_buffer is NULL from
being unloaded; callers are already forced to find a new buffer then)
These were the main causes of b_nwindows bugs, as these cases could not be
reliably detected, and went largely unhandled.
NOTE: if close_buffer() autocmds switch buffers, close_buffer() is not called
for that new buffer before decrementing b_nwindows. This may skip side-effects
like from 'bufhidden', but I think it's mostly harmless, and was already
happening in other places.
Let's see how this goes... Other details: (I have lots to say!)
It's OK to pass a win to close_buffer() that isn't showing buf (used by
set_curbuf()). In that case, we skip some side-effects and don't decrement
b_nwindows, but may still unload buf if hidden.
buf_freeall() now returns whether it freed anything. Removes some repeated
checks in close_buffer().
Preserve close_buffer()'s behaviour when called by win_free_popup() after its
popup was already removed from the window list. This made win_valid_any_tab()
return FALSE, so we skip things that originally checked it in that case.
Add "set_context" to close_buffer() to preserve do_ecmd()'s behaviour of only
setting b_last_cursor and/or calling buflist_setfpos() when not splitting
(see 7.2.041: https://groups.google.com/g/vim_dev/c/ZGgNvaylNzI/m/WHxjhnuxqB0J)
Without this, Test_marks_cmd() fails from its ' mark differing. Don't use
oldwin though; it's not always the window with the closed buf, especially
after BufLeave autocmds in do_ecmd(). Also, only set context if win is really
displaying buf.
Don't bail in do_ecmd() if buf was deleted but curwin->w_buffer is NULL; that
leaves curwin open to a NULL buffer! Use lastbuf instead, like set_curbuf().
I don't think it's possible for buf to be deleted by close_buffer() anyway, as
b_locked was set (which I can't see a way to bypass, unlike b_locked_split).
Maybe such checks can be removed, but I'd rather not risk that here.
Don't set curwin to previouswin in set_curbuf(); shouldn't be needed, otherwise
may lead to curbuf != curwin->w_buffer if autocmds switch to a window showing
buf, as that skips enter_buffer()? Was introduced back in 7.3.557 to avoid
cases where autocmds switch windows, possibly leaving previouswin with a NULL
buffer. Since 7.4.2312 and 7.4.2328, close_buffer() and buf_freeall() already
handles this. I've added an assert() as a sanity check anyway.
In free_all_mem(), set b_nwindows to 0 before close_buffer() so buffers can be
wiped if still in a window before win_free_all(). Needed as close_buffer() now
skips unloading buffers that aren't hidden if win is NULL. If it's possible for
free_all_mem()'s :tabonly! and :only! to not close all windows before freeing,
then this issue was also previously possible if b_nwindows > 1.
related: #19728
Signed-off-by: Sean Dewar <6256228+seandewar@users.noreply.github.com>
Signed-off-by: Christian Brabandt <cb@256bit.org>
0 commit comments