Gdiff: restore cursor position in original window

[blueyed]

Fixes #428

[tpope]

Find a way to do this in the diff pane please. #36 explains why I switched the focus.

[blueyed]

@tpope
While doing it in the diff pane could work somehow maybe, I find it more useful to stay in the original window, at the original position, instead of getting somewhere nearby (or thrown at the top like it's currently) in the diff pane.

#36 covers multiple issues: e.g. "stuck in diff mode" is not a problem with this pull request, as far as I can see.

[tpope]

It is, actually. wincmd p can fail when other plugins are up to shenanigans on certain autocmd events. But the actual reason I changed the default is because it's awkward to close the diff window if you can't use <C-W>o (also covered in that big sprawly issue).

Losing the current line is unfortunate, and it's a tradeoff I deliberated for months before making. But I'm sticking to it. If you can find a way to restore it without regressing on window focus, great. You might look at winline().

[tpope]

There's also 'cursorbind', but that's relatively new, so you'll need to cope gracefully if it's missing.

[blueyed]

Please consider re-opening this PR, so that we can discuss this in the open field.

wincmd p can fail when other plugins are up to shenanigans on certain autocmd events.
Wouldn't nocautocmd help here?

But the actual reason I changed the default is because it's awkward to close the diff window if you can't use <C-W>o (also covered in that big sprawly issue).
I have a function/map for this. From inside of fugitive, a Gdiffoff command might be easier probably:

" Close any corresponding fugitive diff buffer.
function! MyCloseDiff()
  if (&diff == 0 || getbufvar('#', '&diff') == 0)
        \ && (bufname('%') !~ '^fugitive:' && bufname('#') !~ '^fugitive:')
    echom "Not in diff view."
    return
  endif

  diffoff " safety net / required to workaround powerline issue

" Close current buffer if alternate is not fugitive but current one is.
  if bufname('#') !~ '^fugitive:' && bufname('%') =~ '^fugitive:'
    if bufwinnr("#") == -1
" XXX: might not work reliable (old comment)
      b #
      bd #
    else
      bd
    endif
  else
    bd #
  endif
endfunction
nnoremap <Leader>gD :call MyCloseDiff()<cr>

You might look at winline().
What is your idea behind it? How is the window line relevant to get to the matching line in the diff buffer?

There's also 'cursorbind'

It is handled in s:diff_restore already, although not being set by fugitive (diff* activates it?!)

if has('cursorbind')
  let restore .= (&l:cursorbind ? ' ' : ' no') . 'cursorbind'
endif

[tpope]

On Wed, Jan 29, 2014 at 9:19 PM, Daniel Hahler notifications@github.comwrote:

wincmd p can fail when other plugins are up to shenanigans on certain
autocmd events.
Wouldn't nocautocmd help here?

Possible but doubtful. The actual problematic command is probably the
split, and you can't disable them there without breaking everything.

You might look at winline().
What is your idea behind it? How is the window line relevant to get to the
matching line in the diff buffer?

Find the winline in the original buffer then go to the same winline in the
other buffer. Doing this probably involves bouncing between buffers so
above caveats apply. Although if you can preserve the use of diffsplit,
then hopefully only the cursor repositioning would fail in this broken
setup.

There's also 'cursorbind'

It is handled in s:diff_restore already, although not being set by
fugitive (diff* activates it?!)

if has('cursorbind')
let restore .= (&l:cursorbind ? ' ' : ' no') . 'cursorbind'
endif

Not sure it's ever being set, but see #122.

[tpope]

I've ripped out all the hacks, as they are of dubious value on modern Vim.