Re: "." fails to repeat the last executed operator with ":normal v"

[glts]

Here are the contents of the redo buffer when "." is pressed, for both
cases (I obtained these with get_inserted()):

1. "d:call Select()\n"
2. "vd"

So, in the first case when we enter "dfoo", "foo" is mapped and put
verbatim in the redo buffer.

In the second case, "dfoo" is somehow translated to "vd". The effect is
the same as for "1vd", meaning: make a Visual selection with the same
size as the previous one, then delete it.

This is strange. I don't understand how the second case works yet but I
will try to find out more when I have time.

glts

[glts]

On Saturday, March 16, 2013 6:37:18 AM UTC+1, Kana Natsuno wrote:
> Don't you have any idea about this problem?

On Saturday, March 16, 2013 12:06:50 PM UTC+1, glts wrote:
> Here are the contents of the redo buffer when ...

The "repeat.vim" plugin was created to solve this problem but it doesn't
work for omaps neither. I don't know if you saw it but there is some
discussion here:

https://github.com/tpope/vim-repeat/issues/8

In my opinion, the redo buffer should be the same in both of your
examples, namely "d:call Select()\n". But since there is only one redo
buffer and you can call functions/:normal recursively, there's always
the chance of it being overwritten.

Folks, would it make sense to add another "top-level" redo buffer? This
would be left untouched by nested :normal commands.

[Kana Natsuno]

On Sun, Mar 17, 2013 at 8:03 PM, glts wrote:
> The "repeat.vim" plugin was created to solve this problem but it doesn't
> work for omaps neither. I don't know if you saw it but there is some
> discussion here:
>
> https://github.com/tpope/vim-repeat/issues/8

I already saw the discussion, because several people asked me about
repeatability of custom text objects implemented with vim-textobj-user
and most of them also told me the link.

But I don't believe that repeat.vim is a right solution for the problem.
repeat.vim is to enable repeatability for custom operators which take
extra arguments such as a {target} for ys{motion}{target} provided by
surround.vim. Repeatability of custom motions and custom text objects
should be covered by another layer. Especially, the "solution" posted
in the discussion

https://github.com/tpope/vim-repeat/issues/8#issuecomment-13951082

does not work with custom operators using repeat.vim.

> In my opinion, the redo buffer should be the same in both of your
> examples, namely "d:call Select()\n". But since there is only one redo
> buffer and you can call functions/:normal recursively, there's always
> the chance of it being overwritten.
>
> Folks, would it make sense to add another "top-level" redo buffer? This
> would be left untouched by nested :normal commands.

Thank you for the summary. It seems to be reasonable for me, but there
might be drawbacks about compatibility or difficulty for further
maintenance. I'd like to hear others' opinions.

[Sung Pae]

On Sun, Mar 17, 2013 at 05:08:29AM -0700, Kana Natsuno wrote:

> Especially, the "solution" posted in the discussion
>
> https://github.com/tpope/vim-repeat/issues/8#issuecomment-13951082
>
> does not work with custom operators using repeat.vim.

While I do hope Vim will gain native support for repeating custom
commands with `.`, I would like to point out that the posted solution,
while hackish, indeed does work with custom operators, custom motions,
and even in combination.

I am close to releasing a plugin with 18 custom text objects and cursor
motions for S-Expressions, and I have a satisfactory¹ implementation of
repeatable operator-pending commands here:

https://github.com/guns/vim-sexp/blob/90cc836e9c83d4b96c6c5e971febc33a5d73f7e7/plugin/sexp.vim#L103..L132

The motion mappings created by the above function work with both native
operators and with the custom operators provided by two other plugins by
Tim Pope: vim-surround and vim-commentary. I don't know offhand if Tim
had to put in any extra work to make this possible, but I don't believe
so.

Sung Pae

¹ Not perfect. The `c` command is special cased for now and the final
position of the cursor is sometimes different than with native
operations, but I think both issues could be resolved with some
ingenuity.

[Sung Pae]

On Tue, Mar 19, 2013 at 05:21:42PM -0500, Sung Pae wrote:
> On Sun, Mar 17, 2013 at 05:08:29AM -0700, Kana Natsuno wrote:
>
> > Especially, the "solution" posted in the discussion
> >
> > https://github.com/tpope/vim-repeat/issues/8#issuecomment-13951082
> >
> > does not work with custom operators using repeat.vim.
>
> While I do hope Vim will gain native support for repeating custom
> commands with `.`

Apologies for replying to myself here, but I wish to emphasize that I
am in _complete_ agreement with Kana Natsuno about his proposal. This
should optimally "just work" without any of the hackery I've posted.

Operators + cursor motions are one of the pillars of the Vim editing
philosophy and fixing this issue would definitely encourage people to
create their own innovative FileType-specific motions.

I would be happy to supply a patch if it is acknowledged as a bug, and
if Kana is not already in the midst of a solution.

Sung Pae

[Kana Natsuno]

On Wed, Mar 20, 2013 at 7:21 AM, Sung Pae wrote:
> On Sun, Mar 17, 2013 at 05:08:29AM -0700, Kana Natsuno wrote:
>
>> Especially, the "solution" posted in the discussion
>>
>> https://github.com/tpope/vim-repeat/issues/8#issuecomment-13951082
>>
>> does not work with custom operators using repeat.vim.
>
> While I do hope Vim will gain native support for repeating custom
> commands with `.`, I would like to point out that the posted solution,
> while hackish, indeed does work with custom operators, custom motions,
> and even in combination.

No. The "solution" does not work in every situation, especially for
a combination of a custom operator and a custom text object that use
repeat.vim.

> I am close to releasing a plugin with 18 custom text objects and cursor
> motions for S-Expressions, and I have a satisfactory¹ implementation of
> repeatable operator-pending commands here:
>
> https://github.com/guns/vim-sexp/blob/90cc836e9c83d4b96c6c5e971febc33a5d73f7e7/plugin/sexp.vim#L103..L132
>
> The motion mappings created by the above function work with both native
> operators and with the custom operators provided by two other plugins by
> Tim Pope: vim-surround and vim-commentary.

Really? Suppose that we invoke a new Vim process with the following
environment:

------------------------------------------------------------

$ git clone git://github.com/tpope/vim-repeat.git
$ cd vim-repeat
$ git checkout a81bef76031ca1c71766b516417480caeb01c932
$ cd ..
$ git clone git://github.com/guns/vim-sexp.git
$ cd vim-surround
$ git checkout 90cc836e9c83d4b96c6c5e971febc33a5d73f7e7
$ cd ..
$ git clone git://github.com/tpope/vim-surround.git
$ cd vim-surround
$ git checkout 02199ea0080d744ec76b79d74ce56d51d25cf7ae
$ cd ..
$ vim -u NONE -i NONE -N -c "
syntax enable |
set runtimepath+=$PWD/vim-repeat,$PWD/vim-sexp,$PWD/vim-surround |
runtime! plugin/sexp.vim plugin/surround.vim
"

------------------------------------------------------------

Then `:setfiletype scheme` and paste the following snippet:

(let ((a b) (c d))
code)
(let ((e f)) more-code)

Suppose that we do

(1) `1ggf(` (move the cursor the second "(" in the first line)
(2) `ysabr` (surround the "(...)" block with "[" and "]"),
(3) `3ggf(` (move the cursor the second "(" in the third line)
(4) `.` (repeat (2), the last edit command)

We'll get the following result:

(let [((a b) (c d))]
code)
(let [((e f))] more-code)

If we do the following steps on the same snippet:

(1) `1ggf(`
(2) `ysafr` (surround the current form with "[" and "]"),
(3) `3ggf(`
(4) `.`

We'll get the following result:

(let [((a b) (c d))]
code)
(let [((e f)) more-]code)

That's why I don't believe that the "solution" is not a right way to
solve this problem. It does not provide repeatability for every
combination.

[Sung Pae]

On Tue, Mar 19, 2013 at 04:39:11PM -0700, Kana Natsuno wrote:

> On Wed, Mar 20, 2013 at 7:21 AM, Sung Pae wrote:
>
> > The motion mappings created by the above function work with both
> > native operators and with the custom operators provided by two other
> > plugins by Tim Pope: vim-surround and vim-commentary.
>
> Really? Suppose that we invoke a new Vim process with the following
> environment:
>
> ------------------------------------------------------------

Yes, this combination does fail due to the implementation of the `ys`
operator. This can be ameliorated, but it would not be pretty.

Other combinations do not fail (like the `gc` vim-commentary operator
with the `af` text object), and the native operators with custom text
objects work well.

Again, this is a hack against the constraints of the bug you have
pointed out, which I am very much in favor of seeing fixed. Do you have
a patch to supply to the list, or should I look into it?

Sung Pae

[Yukihiro Nakadaira]

Perhaps this works?

onoremap foo :<C-u>call Select()<CR>
function! Select()
  execute "normal! viw\<Esc>"
endfunction

--
Yukihiro Nakadaira - yukihiro....@gmail.com

[Yukihiro Nakadaira]

On Wednesday, March 20, 2013 10:49:19 PM UTC+9, Yukihiro Nakadaira wrote:
> Perhaps this works?
>
> onoremap foo :<C-u>call Select()<CR>
> function! Select()
>   execute "normal! viw\<Esc>"
> endfunction

Oops, this doesn't work. Sorry for noise.

[glts]

Hi

I spent some time looking into this. Attached is a first *proposal* for
a patch for the current Vim version 7.3.875, which -- I believe -- fixes
this.

However, the code surrounding Visual mode and operators is rather ...
tricky to say the least, so I urge you to look it over carefully, and
comment on whether you think it is correct or how it could be improved.

There has not been any response on the status of this issue. Operators
and text objects are two of the major advantages Vim has over other
editors. So I believe it is essential to let users leverage the full
power of their own custom operators and text objects by making them
repeatable with the dot command.

David Bürgin

[Sung Pae]

On Sat, Mar 23, 2013 at 09:56:58AM -0700, glts wrote:

> I spent some time looking into this. Attached is a first *proposal*
> for a patch for the current Vim version 7.3.875, which -- I believe --
> fixes this.

I can confirm it works for simple repeats of custom motions with `d` and
`c` operators, as well the custom `ys` operator that Kana showed was
problematic.

> However, the code surrounding Visual mode and operators is rather ...
> tricky to say the least, so I urge you to look it over carefully,
> and comment on whether you think it is correct or how it could be
> improved.

I see two issues on first glance:

* Counts are not repeated; d2x (x being a custom motion) repeats as
dx only
* Making a visual selection with a custom operator clobbers the redo

It's promising that your small patch enables so much of the desired
functionality. Please continue!

> There has not been any response on the status of this issue. Operators
> and text objects are two of the major advantages Vim has over other
> editors. So I believe it is essential to let users leverage the full
> power of their own custom operators and text objects by making them
> repeatable with the dot command.

Perhaps we will be able to convince Bram if the solution is clean and
unobtrusive.

Sung Pae

[glts]

Sung Pae, thank you for your feedback.

On Sun, Mar 24, 2013 at 12:40 AM, Sung Pae <se...@sungpae.com> wrote:
> I see two issues on first glance:
>
> * Counts are not repeated; d2x (x being a custom motion) repeats as
> dx only

Yes, but this is due to another bug. My patch does not affect this.

The v:count variables are treated differently in omaps. Compare:

:nnoremap <silent> x :<C-U>exe "norm! d".v:count1."e"<CR>
:onoremap <silent> x :<C-U>exe "norm! ".v:count1."e"<CR>

If you try the normal mode mapping, e.g. "2x", then the dot command will
have the same effect as "2x". You can also override the count by giving
a count to the dot command, e.g. "3.". This is the expected behaviour.

The omap on the other hand doesn't work like that. Try "d2x" and then
repeat with the dot command: The original count is not used. You have to
use "2." to get the same as with "d2x".

This is a different issue, and I'd rather put it off for the moment.

> * Making a visual selection with a custom operator clobbers the redo

Can you give an example? I don't see the problem.

[Sung Pae]

On Sun, Mar 24, 2013 at 06:43:47PM +0100, glts wrote:

> On Sun, Mar 24, 2013 at 12:40 AM, Sung Pae <se...@sungpae.com> wrote:
> > I see two issues on first glance:
> >
> > * Counts are not repeated; d2x (x being a custom motion) repeats as
> > dx only
>
> Yes, but this is due to another bug. My patch does not affect this.
>
> The v:count variables are treated differently in omaps. Compare:

Ah yes, you're totally correct. I remember now that I had to work
around that myself by saving v:count to a buffer-local variable on omap
invocation, then string-replace the v:count variable in the omap rhs
with the buffer-local "b:_count", so that the cached value would be used
on repeat. It was very nasty, but it worked.

> This is a different issue, and I'd rather put it off for the moment.

Yes, I imagine a new count variable needs to be created to handle omaps.

> > * Making a visual selection with a custom operator clobbers the redo
>
> Can you give an example? I don't see the problem.

I apologize, that was a bit vague.

Let's create a simple (and incomplete) custom motion:

function! SelectInCaps()
let [bl, bc] = searchpos('\u', 'cbW')
let [el, ec] = searchpos('.\u\ze', 'W')
call setpos("'<", [0, bl, bc, 0])
call setpos("'>", [0, el, ec, 0])
normal! gv
endfunction

Then map it to both visual and operator-pending modes:

vmap ic :<C-U>call SelectInCaps()<CR>
omap ic :<C-U>call SelectInCaps()<CR>

Repeating the command dic with the . command works now with your patch:

Foo|BarBazQuux
Foo|BazQuux " dic
Foo|Quuz " .

However, if we invoke vic after running dic, the . command no longer has
any affect:

Foo|BarBazQuux
Foo|BazQuux " dic
Foo|BazQuux " vic<Esc>
Foo|BazQuux " .

This is not true of the builtin motions like `aw`; visual commands do
not clobber the repeat buffer (or however it may work).

Thank you for your work on this!

Sung Pae

[glts]

On Mon, Mar 25, 2013 at 2:08 AM, Sung Pae <se...@sungpae.com> wrote:
> Let's create a simple (and incomplete) custom motion:
>
> function! SelectInCaps()
> let [bl, bc] = searchpos('\u', 'cbW')
> let [el, ec] = searchpos('.\u\ze', 'W')
> call setpos("'<", [0, bl, bc, 0])
> call setpos("'>", [0, el, ec, 0])
> normal! gv
> endfunction
>
> Then map it to both visual and operator-pending modes:
>
> vmap ic :<C-U>call SelectInCaps()<CR>
> omap ic :<C-U>call SelectInCaps()<CR>
>
> Repeating the command dic with the . command works now with your patch:
>
> Foo|BarBazQuux
> Foo|BazQuux " dic
> Foo|Quuz " .
>
> However, if we invoke vic after running dic, the . command no longer has
> any affect:
>
> Foo|BarBazQuux
> Foo|BazQuux " dic
> Foo|BazQuux " vic<Esc>
> Foo|BazQuux " .

Yes, sorry, and thanks for reporting. Operator and Visual mode handling
isn't easy and I'm learning it the hard way ...

Anyway, after another session in the debugger I came up with the slight
amendment in the attachment. If you would like to try it, be my guest. I
hope I can find the time to write a few tests.

David

[Sung Pae]

On Mon, Mar 25, 2013 at 10:14:48PM +0100, glts wrote:

> On Mon, Mar 25, 2013 at 2:08 AM, Sung Pae <se...@sungpae.com> wrote:
>

> > Let's create a simple (and incomplete) custom motion: …

>
> Anyway, after another session in the debugger I came up with the
> slight amendment in the attachment. If you would like to try it, be my
> guest. I hope I can find the time to write a few tests.

Yes, this definitely solves the problem with visual selections clearing
the repeat. I tried poking around a bit more, and I couldn't find any
obvious problems, which is exciting.

All that appears to be left to achieve parity with builtin cursor
motions is the issue with v:count. I will report any problems should I
find them.

Thank you for looking into this.

Sung Pae