updated nrrwrgn integration

This commit is contained in:
Christian Brabandt 2014-05-03 14:16:50 +02:00
parent db0ea7a970
commit 534d8e844b

View File

@ -8,13 +8,43 @@ endif
function! airline#extensions#nrrwrgn#apply(...)
if exists(":WidenRegion") == 2
let spc = g:airline_symbols.space
call a:1.add_section('airline_a', printf('%s[Narrowed%s#%d]', spc, spc, b:nrrw_instn))
if exists('*airline#extensions#branch#get_head')
call a:1.add_section('airline_b', spc.'%{airline#extensions#branch#get_head()}'.spc)
if !exists("*nrrwrgn#NrrwRgnStatus()") || empty(nrrwrgn#NrrwRgnStatus())
call a:1.add_section('airline_a', printf('%s[Narrowed%s#%d]', spc, spc, b:nrrw_instn))
let bufname=(get(b:, 'orig_buf', 0) ? bufname(b:orig_buf) : substitute(bufname('%'), '^Nrrwrgn_\zs.*\ze_\d\+$', submatch(0), ''))
call a:1.add_section('airline_c', spc.bufname.spc)
call a:1.split()
else
let dict=nrrwrgn#NrrwRgnStatus()
let vmode = { 'v': 'Char ', 'V': 'Line ', '': 'Block '}
let mode = dict.visual ? vmode[dict.visual] : vmode['V']
let winwidth = winwidth(0)
if winwidth < 80
let mode = mode[0]
endif
let title = (winwidth < 80 ? "Nrrw" : "Narrowed ")
let multi = (winwidth < 80 ? 'M' : 'Multi')
call a:1.add_section('airline_a', printf('[%s%s%s#%d]%s', (dict.multi ? multi : ""),
\ title, mode, b:nrrw_instn, spc))
let name = dict.fullname
if name !=# '[No Name]'
if winwidth > 100
" need some space
let name = fnamemodify(dict.fullname, ':~')
if strlen(name) > 8
" shorten name
let name = substitute(name, '\(.\)[^/\\]*\([/\\]\)', '\1\2', 'g')
endif
else
let name = fnamemodify(dict.fullname, ':t')
endif
endif
let range=(dict.multi ? '' : printf("[%d-%d]", dict.start[1], dict.end[1]))
call a:1.add_section('airline_c', printf("%s %s %s", name, range, dict.enabled ? "\u2713" : '!'))
call a:1.split()
call a:1.add_section('airline_x', get(g:, 'airline_section_x').spc)
call a:1.add_section('airline_y', spc.get(g:, 'airline_section_y').spc)
call a:1.add_section('airline_z', spc.get(g:, 'airline_section_z'))
endif
let bufname=(get(b:, 'orig_buf', 0) ? bufname(b:orig_buf) : substitute(bufname('%'), '^Nrrwrgn_\zs.*\ze_\d\+$', submatch(0), ''))
call a:1.add_section('airline_c', spc.bufname.spc)
call a:1.split()
return 1
endif
endfunction