neovim/test/old/testdir/test_arglist.vim

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

752 lines
18 KiB
VimL
Raw Normal View History

" Test argument list commands
source check.vim
source shared.vim
source term_util.vim
func Reset_arglist()
args a | %argd
endfunc
func Test_argidx()
args a b c
last
call assert_equal(2, argidx())
%argdelete
call assert_equal(0, argidx())
" doing it again doesn't result in an error
%argdelete
call assert_equal(0, argidx())
call assert_fails('2argdelete', 'E16:')
args a b c
call assert_equal(0, argidx())
next
call assert_equal(1, argidx())
next
call assert_equal(2, argidx())
1argdelete
call assert_equal(1, argidx())
1argdelete
call assert_equal(0, argidx())
1argdelete
call assert_equal(0, argidx())
endfunc
func Test_argadd()
call Reset_arglist()
%argdelete
argadd a b c
call assert_equal(0, argidx())
%argdelete
argadd a
call assert_equal(0, argidx())
argadd b c d
call assert_equal(0, argidx())
call Init_abc()
argadd x
call Assert_argc(['a', 'b', 'x', 'c'])
call assert_equal(1, argidx())
call Init_abc()
0argadd x
call Assert_argc(['x', 'a', 'b', 'c'])
call assert_equal(2, argidx())
call Init_abc()
1argadd x
call Assert_argc(['a', 'x', 'b', 'c'])
call assert_equal(2, argidx())
call Init_abc()
$argadd x
call Assert_argc(['a', 'b', 'c', 'x'])
call assert_equal(1, argidx())
call Init_abc()
$argadd x
+2argadd y
call Assert_argc(['a', 'b', 'c', 'x', 'y'])
call assert_equal(1, argidx())
%argd
edit d
arga
call assert_equal(1, len(argv()))
call assert_equal('d', get(argv(), 0, ''))
%argd
edit some\ file
arga
call assert_equal(1, len(argv()))
call assert_equal('some file', get(argv(), 0, ''))
%argd
new
arga
call assert_equal(0, len(argv()))
if has('unix')
call assert_fails('argadd `Xdoes_not_exist`', 'E479:')
endif
endfunc
func Test_argadd_empty_curbuf()
new
let curbuf = bufnr('%')
call writefile(['test', 'Xargadd'], 'Xargadd')
" must not re-use the current buffer.
argadd Xargadd
call assert_equal(curbuf, bufnr('%'))
call assert_equal('', bufname('%'))
call assert_equal(1, '$'->line())
rew
call assert_notequal(curbuf, '%'->bufnr())
call assert_equal('Xargadd', '%'->bufname())
call assert_equal(2, line('$'))
%argd
bwipe!
endfunc
func Init_abc()
args a b c
next
endfunc
func Assert_argc(l)
call assert_equal(len(a:l), argc())
let i = 0
while i < len(a:l) && i < argc()
call assert_equal(a:l[i], argv(i))
let i += 1
endwhile
endfunc
" Test for [count]argument and [count]argdelete commands
" Ported from the test_argument_count.in test script
func Test_argument()
call Reset_arglist()
let save_hidden = &hidden
set hidden
let g:buffers = []
augroup TEST
au BufEnter * call add(buffers, expand('%:t'))
augroup END
argadd a b c d
$argu
$-argu
-argu
1argu
+2argu
augroup TEST
au!
augroup END
call assert_equal(['d', 'c', 'b', 'a', 'c'], g:buffers)
call assert_equal("\na b [c] d ", execute(':args'))
.argd
call assert_equal(['a', 'b', 'd'], argv())
-argd
call assert_equal(['a', 'd'], argv())
$argd
call assert_equal(['a'], argv())
1arga c
1arga b
$argu
$arga x
call assert_equal(['a', 'b', 'c', 'x'], argv())
0arga y
call assert_equal(['y', 'a', 'b', 'c', 'x'], argv())
%argd
call assert_equal([], argv())
arga a b c d e f
2,$-argd
call assert_equal(['a', 'f'], argv())
let &hidden = save_hidden
let save_columns = &columns
let &columns = 79
try
exe 'args ' .. join(range(1, 81))
call assert_equal(join([
\ '',
\ '[1] 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 ',
\ '2 7 12 17 22 27 32 37 42 47 52 57 62 67 72 77 ',
\ '3 8 13 18 23 28 33 38 43 48 53 58 63 68 73 78 ',
\ '4 9 14 19 24 29 34 39 44 49 54 59 64 69 74 79 ',
\ '5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 ',
\ ], "\n"),
\ execute('args'))
" No trailing newline with one item per row.
let long_arg = repeat('X', 81)
exe 'args ' .. long_arg
call assert_equal("\n[".long_arg.']', execute('args'))
finally
let &columns = save_columns
endtry
" Setting argument list should fail when the current buffer has unsaved
" changes
%argd
enew!
set modified
call assert_fails('args x y z', 'E37:')
args! x y z
call assert_equal(['x', 'y', 'z'], argv())
call assert_equal('x', expand('%:t'))
last | enew | argu
call assert_equal('z', expand('%:t'))
%argdelete
call assert_fails('argument', 'E163:')
endfunc
func Test_list_arguments()
" Clean the argument list
arga a | %argd
" four args half the screen width makes two lines with two columns
let aarg = repeat('a', &columns / 2 - 4)
let barg = repeat('b', &columns / 2 - 4)
let carg = repeat('c', &columns / 2 - 4)
let darg = repeat('d', &columns / 2 - 4)
exe 'argadd ' aarg barg carg darg
redir => result
args
redir END
call assert_match('\[' . aarg . '] \+' . carg . '\n' . barg . ' \+' . darg, trim(result))
" if one arg is longer than half the screen make one column
exe 'argdel' aarg
let aarg = repeat('a', &columns / 2 + 2)
exe '0argadd' aarg
redir => result
args
redir END
call assert_match(aarg . '\n\[' . barg . ']\n' . carg . '\n' . darg, trim(result))
%argdelete
endfunc
func Test_args_with_quote()
" Only on Unix can a file name include a double quote.
if has('unix')
args \"foobar
call assert_equal('"foobar', argv(0))
%argdelete
endif
endfunc
" Test for 0argadd and 0argedit
" Ported from the test_argument_0count.in test script
func Test_zero_argadd()
call Reset_arglist()
arga a b c d
2argu
0arga added
call assert_equal(['added', 'a', 'b', 'c', 'd'], argv())
2argu
arga third
call assert_equal(['added', 'a', 'third', 'b', 'c', 'd'], argv())
%argd
arga a b c d
2argu
0arge edited
call assert_equal(['edited', 'a', 'b', 'c', 'd'], argv())
2argu
arga third
call assert_equal(['edited', 'a', 'third', 'b', 'c', 'd'], argv())
2argu
argedit file\ with\ spaces another file
call assert_equal(['edited', 'a', 'file with spaces', 'another', 'file', 'third', 'b', 'c', 'd'], argv())
call assert_equal('file with spaces', expand('%'))
endfunc
" Test for argc()
func Test_argc()
call Reset_arglist()
call assert_equal(0, argc())
argadd a b
call assert_equal(2, argc())
endfunc
" Test for arglistid()
func Test_arglistid()
call Reset_arglist()
arga a b
call assert_equal(0, arglistid())
split
arglocal
call assert_equal(1, arglistid())
tabnew | tabfirst
call assert_equal(0, arglistid(2))
call assert_equal(1, arglistid(1, 1))
call assert_equal(0, arglistid(2, 1))
call assert_equal(1, arglistid(1, 2))
tabonly | only | enew!
argglobal
call assert_equal(0, arglistid())
endfunc
" Tests for argv() and argc()
func Test_argv()
call Reset_arglist()
call assert_equal([], argv())
call assert_equal("", argv(2))
call assert_equal(0, argc())
argadd a b c d
call assert_equal(4, argc())
call assert_equal('c', argv(2))
let w1_id = win_getid()
split
let w2_id = win_getid()
arglocal
args e f g
tabnew
let w3_id = win_getid()
split
let w4_id = win_getid()
argglobal
tabfirst
call assert_equal(4, argc(w1_id))
call assert_equal('b', argv(1, w1_id))
call assert_equal(['a', 'b', 'c', 'd'], argv(-1, w1_id))
call assert_equal(3, argc(w2_id))
call assert_equal('f', argv(1, w2_id))
call assert_equal(['e', 'f', 'g'], argv(-1, w2_id))
call assert_equal(3, argc(w3_id))
call assert_equal('e', argv(0, w3_id))
call assert_equal(['e', 'f', 'g'], argv(-1, w3_id))
call assert_equal(4, argc(w4_id))
call assert_equal('c', argv(2, w4_id))
call assert_equal(['a', 'b', 'c', 'd'], argv(-1, w4_id))
call assert_equal(4, argc(-1))
call assert_equal(3, argc())
call assert_equal('d', argv(3, -1))
call assert_equal(['a', 'b', 'c', 'd'], argv(-1, -1))
tabonly | only | enew!
" Negative test cases
call assert_equal(-1, argc(100))
call assert_equal('', argv(1, 100))
call assert_equal([], argv(-1, 100))
call assert_equal('', argv(10, -1))
endfunc
" Test for the :argedit command
func Test_argedit()
call Reset_arglist()
argedit a
call assert_equal(['a'], argv())
call assert_equal('a', expand('%:t'))
argedit b
call assert_equal(['a', 'b'], argv())
call assert_equal('b', expand('%:t'))
argedit a
call assert_equal(['a', 'b', 'a'], argv())
call assert_equal('a', expand('%:t'))
" When file name case is ignored, an existing buffer with only case
" difference is re-used.
argedit C D
call assert_equal('C', expand('%:t'))
call assert_equal(['a', 'b', 'a', 'C', 'D'], argv())
argedit c
if has('fname_case')
call assert_equal(['a', 'b', 'a', 'C', 'c', 'D'], argv())
else
call assert_equal(['a', 'b', 'a', 'C', 'C', 'D'], argv())
endif
0argedit x
if has('fname_case')
call assert_equal(['x', 'a', 'b', 'a', 'C', 'c', 'D'], argv())
else
call assert_equal(['x', 'a', 'b', 'a', 'C', 'C', 'D'], argv())
endif
enew! | set modified
call assert_fails('argedit y', 'E37:')
argedit! y
if has('fname_case')
call assert_equal(['x', 'y', 'y', 'a', 'b', 'a', 'C', 'c', 'D'], argv())
else
call assert_equal(['x', 'y', 'y', 'a', 'b', 'a', 'C', 'C', 'D'], argv())
endif
%argd
bwipe! C
bwipe! D
" :argedit reuses the current buffer if it is empty
%argd
" make sure to use a new buffer number for x when it is loaded
bw! x
new
let a = bufnr()
argedit x
call assert_equal(a, bufnr())
call assert_equal('x', bufname())
%argd
bw! x
endfunc
" Test for the :argdedupe command
func Test_argdedupe()
call Reset_arglist()
argdedupe
call assert_equal([], argv())
args a a a aa b b a b aa
argdedupe
call assert_equal(['a', 'aa', 'b'], argv())
args a b c
argdedupe
call assert_equal(['a', 'b', 'c'], argv())
args a
argdedupe
call assert_equal(['a'], argv())
args a A b B
argdedupe
if has('fname_case')
call assert_equal(['a', 'A', 'b', 'B'], argv())
else
call assert_equal(['a', 'b'], argv())
endif
args a b a c a b
last
argdedupe
next
call assert_equal('c', expand('%:t'))
args a ./a
argdedupe
call assert_equal(['a'], argv())
%argd
endfunc
" Test for the :argdelete command
func Test_argdelete()
call Reset_arglist()
args aa a aaa b bb
argdelete a*
call assert_equal(['b', 'bb'], argv())
call assert_equal('aa', expand('%:t'))
last
argdelete %
call assert_equal(['b'], argv())
call assert_fails('argdelete', 'E610:')
call assert_fails('1,100argdelete', 'E16:')
call assert_fails('argdel /\)/', 'E55:')
call assert_fails('1argdel 1', 'E474:')
call Reset_arglist()
args a b c d
next
argdel
call Assert_argc(['a', 'c', 'd'])
%argdel
call assert_fails('argdel does_not_exist', 'E480:')
endfunc
func Test_argdelete_completion()
args foo bar
call feedkeys(":argdelete \<C-A>\<C-B>\"\<CR>", 'tx')
call assert_equal('"argdelete bar foo', @:)
call feedkeys(":argdelete x \<C-A>\<C-B>\"\<CR>", 'tx')
call assert_equal('"argdelete x bar foo', @:)
%argd
endfunc
" Tests for the :next, :prev, :first, :last, :rewind commands
func Test_argpos()
call Reset_arglist()
args a b c d
last
call assert_equal(3, argidx())
call assert_fails('next', 'E165:')
prev
call assert_equal(2, argidx())
Next
call assert_equal(1, argidx())
first
call assert_equal(0, argidx())
call assert_fails('prev', 'E164:')
3next
call assert_equal(3, argidx())
rewind
call assert_equal(0, argidx())
%argd
endfunc
" Test for autocommand that redefines the argument list, when doing ":all".
func Test_arglist_autocmd()
autocmd BufReadPost Xxx2 next Xxx2 Xxx1
call writefile(['test file Xxx1'], 'Xxx1')
call writefile(['test file Xxx2'], 'Xxx2')
call writefile(['test file Xxx3'], 'Xxx3')
new
" redefine arglist; go to Xxx1
next! Xxx1 Xxx2 Xxx3
" open window for all args; Reading Xxx2 will try to change the arglist and
" that will fail
call assert_fails("all", "E1156:")
call assert_equal('test file Xxx1', getline(1))
wincmd w
call assert_equal('test file Xxx2', getline(1))
wincmd w
call assert_equal('test file Xxx3', getline(1))
autocmd! BufReadPost Xxx2
enew! | only
call delete('Xxx1')
call delete('Xxx2')
call delete('Xxx3')
argdelete Xxx*
bwipe! Xxx1 Xxx2 Xxx3
endfunc
func Test_arg_all_expand()
call writefile(['test file Xxx1'], 'Xx x')
next notexist Xx\ x runtest.vim
call assert_equal('notexist Xx\ x runtest.vim', expand('##'))
call delete('Xx x')
endfunc
func Test_large_arg()
" Argument longer or equal to the number of columns used to cause
" access to invalid memory.
exe 'argadd ' .repeat('x', &columns)
args
endfunc
func Test_argdo()
next! Xa.c Xb.c Xc.c
new
let l = []
argdo call add(l, expand('%'))
call assert_equal(['Xa.c', 'Xb.c', 'Xc.c'], l)
bwipe Xa.c Xb.c Xc.c
endfunc
" Test for quitting Vim with unedited files in the argument list
func Test_quit_with_arglist()
CheckRunVimInTerminal
let buf = RunVimInTerminal('', {'rows': 6})
call term_sendkeys(buf, ":set nomore\n")
call term_sendkeys(buf, ":args a b c\n")
call term_sendkeys(buf, ":quit\n")
call TermWait(buf)
call WaitForAssert({-> assert_match('^E173:', term_getline(buf, 6))})
call StopVimInTerminal(buf)
" Try :confirm quit with unedited files in arglist
let buf = RunVimInTerminal('', {'rows': 6})
call term_sendkeys(buf, ":set nomore\n")
call term_sendkeys(buf, ":args a b c\n")
call term_sendkeys(buf, ":confirm quit\n")
call TermWait(buf)
call WaitForAssert({-> assert_match('^\[Y\]es, (N)o: *$',
\ term_getline(buf, 6))})
call term_sendkeys(buf, "N")
call TermWait(buf)
call term_sendkeys(buf, ":confirm quit\n")
call WaitForAssert({-> assert_match('^\[Y\]es, (N)o: *$',
\ term_getline(buf, 6))})
call term_sendkeys(buf, "Y")
call TermWait(buf)
call WaitForAssert({-> assert_equal("finished", term_getstatus(buf))})
only!
" When this test fails, swap files are left behind which breaks subsequent
" tests
call delete('.a.swp')
call delete('.b.swp')
call delete('.c.swp')
endfunc
" Test for ":all" not working when in the cmdline window
func Test_all_not_allowed_from_cmdwin()
au BufEnter * all
next x
" Use try/catch here, somehow assert_fails() doesn't work on MS-Windows
" console.
let caught = 'no'
try
exe ":norm! 7q?apat\<CR>"
catch /E11:/
let caught = 'yes'
endtry
call assert_equal('yes', caught)
au! BufEnter
endfunc
func Test_clear_arglist_in_all()
n 0 00 000 0000 00000 000000
au WinNew 0 n 0
call assert_fails("all", "E1156")
au! *
endfunc
" Test for the :all command
func Test_all_command()
%argdelete
" :all command should not close windows with files in the argument list,
" but can rearrange the windows.
args Xargnew1 Xargnew2
%bw!
edit Xargold1
split Xargnew1
let Xargnew1_winid = win_getid()
split Xargold2
split Xargnew2
let Xargnew2_winid = win_getid()
split Xargold3
all
call assert_equal(2, winnr('$'))
call assert_equal([Xargnew1_winid, Xargnew2_winid],
\ [win_getid(1), win_getid(2)])
call assert_equal([bufnr('Xargnew1'), bufnr('Xargnew2')],
\ [winbufnr(1), winbufnr(2)])
" :all command should close windows for files which are not in the
" argument list in the current tab page.
%bw!
edit Xargold1
split Xargold2
tabedit Xargold3
split Xargold4
tabedit Xargold5
tabfirst
all
call assert_equal(3, tabpagenr('$'))
call assert_equal([bufnr('Xargnew1'), bufnr('Xargnew2')], tabpagebuflist(1))
call assert_equal([bufnr('Xargold4'), bufnr('Xargold3')], tabpagebuflist(2))
call assert_equal([bufnr('Xargold5')], tabpagebuflist(3))
" :tab all command should close windows for files which are not in the
" argument list across all the tab pages.
%bw!
edit Xargold1
split Xargold2
tabedit Xargold3
split Xargold4
tabedit Xargold5
tabfirst
args Xargnew1 Xargnew2
tab all
call assert_equal(2, tabpagenr('$'))
call assert_equal([bufnr('Xargnew1')], tabpagebuflist(1))
call assert_equal([bufnr('Xargnew2')], tabpagebuflist(2))
" If a count is specified, then :all should open only that many windows.
%bw!
args Xargnew1 Xargnew2 Xargnew3 Xargnew4 Xargnew5
all 3
call assert_equal(3, winnr('$'))
call assert_equal([bufnr('Xargnew1'), bufnr('Xargnew2'), bufnr('Xargnew3')],
\ [winbufnr(1), winbufnr(2), winbufnr(3)])
" The :all command should not open more than 'tabpagemax' tab pages.
" If there are more files, then they should be opened in the last tab page.
%bw!
set tabpagemax=3
tab all
call assert_equal(3, tabpagenr('$'))
call assert_equal([bufnr('Xargnew1')], tabpagebuflist(1))
call assert_equal([bufnr('Xargnew2')], tabpagebuflist(2))
call assert_equal([bufnr('Xargnew3'), bufnr('Xargnew4'), bufnr('Xargnew5')],
\ tabpagebuflist(3))
set tabpagemax&
" Without the 'hidden' option, modified buffers should not be closed.
args Xargnew1 Xargnew2
%bw!
edit Xargtemp1
call setline(1, 'temp buffer 1')
split Xargtemp2
call setline(1, 'temp buffer 2')
all
call assert_equal(4, winnr('$'))
call assert_equal([bufnr('Xargtemp2'), bufnr('Xargtemp1'), bufnr('Xargnew1'),
\ bufnr('Xargnew2')],
\ [winbufnr(1), winbufnr(2), winbufnr(3), winbufnr(4)])
" With the 'hidden' option set, both modified and unmodified buffers in
" closed windows should be hidden.
set hidden
all
call assert_equal(2, winnr('$'))
call assert_equal([bufnr('Xargnew1'), bufnr('Xargnew2')],
\ [winbufnr(1), winbufnr(2)])
call assert_equal([1, 1, 0, 0], [getbufinfo('Xargtemp1')[0].hidden,
\ getbufinfo('Xargtemp2')[0].hidden,
\ getbufinfo('Xargnew1')[0].hidden,
\ getbufinfo('Xargnew2')[0].hidden])
set nohidden
" When 'winheight' is set to a large value, :all should open only one
" window.
args Xargnew1 Xargnew2 Xargnew3 Xargnew4 Xargnew5
%bw!
set winheight=9999
call assert_fails('all', 'E36:')
call assert_equal([1, bufnr('Xargnew1')], [winnr('$'), winbufnr(1)])
set winheight&
" When 'winwidth' is set to a large value, :vert all should open only one
" window.
%bw!
set winwidth=9999
call assert_fails('vert all', 'E36:')
call assert_equal([1, bufnr('Xargnew1')], [winnr('$'), winbufnr(1)])
set winwidth&
" empty argument list tests
%bw!
%argdelete
call assert_equal('', execute('args'))
all
call assert_equal(1, winnr('$'))
%argdelete
%bw!
endfunc
" vim: shiftwidth=2 sts=2 expandtab