Re: Bug/non-determinism in output of maparg() and map commands

2017-05-27 18:02 GMT+03:00 Brett Stahlman <brettstahlman@gmail.com>:
> On Sat, May 27, 2017 at 8:35 AM, Nikolay Aleksandrovich Pavlov
> <zyx.vim@gmail.com> wrote:
>> 2017-05-27 12:45 GMT+03:00 Bram Moolenaar <Bram@moolenaar.net>:
>>>
>>> Nikolay Pavlov wrote:
>>>
>>>> 2017-05-26 20:43 GMT+03:00 Bram Moolenaar <Bram@moolenaar.net>:
>>>> >
>>>> > Brett Stahlman wrote:
>>>> >
>>>> >> >> On Tuesday, May 23, 2017 at 8:25:33 AM UTC-5, Brett Stahlman wrote:
>>>> >> >> > On Tue, May 23, 2017 at 4:35 AM, Bram Moolenaar <Bram@moolenaar.net> wrote:
>>>> >> >> > >
>>>> >> >> > > Brett Stahlman wrote:
>>>> >> >> > >
>>>> >> >> %--snip--%
>>>> >> >> > >
>>>> >> >> > > The best solution is probably to also add the raw rhs, with the terminal
>>>> >> >> > > codes replaced. This won't work when changing the terminal type, but
>>>> >> >> > > that is very unlikely to happen.
>>>> >> >> >
>>>> >> >> > You mean adding a key such as "raw_rhs" to the dictionary returned by
>>>> >> >> > maparg()? If so, then yes this would help, but there would still need to
>>>> >> >> > be a way to determine lhs, which is currently even more ambiguous than
>>>> >> >> > rhs. While it's true that I probably already have lhs if I'm calling
>>>> >> >> > maparg(), I need a way to determine which lhs(s) is/are ambiguous with a
>>>> >> >> > given lhs. Mapcheck() gives me only the rhs of the conflicting map. To
>>>> >> >> > save and restore, I'd need to know the lhs in canonical form as well.
>>>> >> >>
>>>> >> >> Perhaps mapcheck() could take an optional arg requesting something more than a simple boolean return. When called with this extra arg, mapcheck() could return a conflicting/ambiguous lhs (or list thereof) in some canonical format (possibly determined by the value of the extra arg itself). As long as the format returned could be fed to maparg(), it would be possible to find conflicting mappings, remove them temporarily, and subsequently restore them...
>>>> >> >
>>>> >> > If you define a mapping you will want to know whether the mapping
>>>> >> > already exists and needs to be restored. For that you can use maparg(),
>>>> >> > no need to use mapcheck().
>>>> >> >
>>>> >> > Not sure why you would want to remove "conflicting" mappings. Perhaps
>>>> >> > when you map the ; key, and the user has ;x mapped? Then you would need
>>>> >> > a list. Adding a maplist() function would be better than adding
>>>> >> > arguments to mapcheck().
>>>> >>
>>>> >> Yes. Very much like that. I'm implementing a sort of transient mode, in
>>>> >> which I'll "shadow" existing maps with very short (generally single
>>>> >> character) mappings, which are expected to be ambiguous/conflicting with
>>>> >> existing maps, and even builtin operators. Of course, when I exit the
>>>> >> transient mode, I'd need to restore the mappings that were shadowed.
>>>> >>
>>>> >> The global and builtin maps are not a problem, since the transient maps use
>>>> >> <buffer> and <nowait>; however, without parsing the output of one of the :map
>>>> >> functions, I have no way of knowing which buf-local mappings will be ambiguous
>>>> >> with the transient maps I'm defining. And parsing the :map output is
>>>> >> problematic for the reasons already mentioned: e.g., no way to tell the
>>>> >> difference between function key <F8> and the corresponding 4 characters. I'd
>>>> >> actually considered taking some sort of iterative approach: e.g., trying all
>>>> >> possible permutations of lhs as input to maparg() and testing the results, in
>>>> >> an attempt to deduce the canonical form, but this would be extremely messy,
>>>> >> and I don't even know whether it would be deterministic... The maplist()
>>>> >> function you mentioned, if it returned all ambiguous left hand sides in
>>>> >> canonical form, or even a list of the corresponding maparg()-style
>>>> >> dictionaries, would be perfect. Of course, there would also need to be a way
>>>> >> to get the rhs's canonical form: e.g., the extra "raw_rhs" key in the maparg()
>>>> >> or maplist() dictionary.
>>>> >
>>>> > OK, so for this you would use maplist() to get the list of mappings to
>>>> > disable, use maparg() to get the current mapping, clear the mapping, do
>>>> > your stuff, then restore the cleared mappings. You then need to make
>>>> > sure you restore the mappings exactly as they were, even when your
>>>> > "stuff" fails miserably.
>>>> >
>>>> > It's a lot easier if we would have a way to temporarily disable
>>>> > mappings. It's mostly the same as above, but you won't need to use
>>>> > maparg() to get the current mapping and the restore operation. Instead
>>>> > you would disable instead of clear, and later re-enable instead of
>>>> > restore. Still need to make sure the re-enbling does happen, no change
>>>> > in that part.
>>>>
>>>> Not sure I understood what exactly you suggest to disable/restore. All
>>>> mappings at once with one command? I would actually disagree here: I
>>>> need something similar for translit3, but it only remaps
>>>> single-character mappings, leaving most of other user mappings alone.
>>>> One mapping at a time? It would be good, but given that request is
>>>> temporary remapping naming the functionality enable/disable looks
>>>> strange. And there are still issues with determining {lhs}.
>>>
>>> Let's use an example: Suppose a plugin has a special mode for entering
>>> data (e.g. chemical formulas). It would then map some keys, e.g. "a".
>>> If the user already has a mapping for "a" it needs to be restored when
>>> leaving the special mode. If the user has mappings starting with "a" we
>>> would like to disable those, to avoid the timeout waiting for the next
>>> character.
>>>
>>> We do not want to disable mappings that don't interfere, to maximise the
>>> freedom for the user to use other mappings at the same time.
>>>
>>>> One of the logical variants would be `:map <push> {lhs}
>>>> {new-rhs}`/`:unmap <push> {lhs}`+`:map <pop> {lhs}`, but this is hard
>>>> to implement and is rather limited, though less limited then
>>>> enable/disable everything variant.
>>>
>>> This quickly gets complicated if we need to take into account all the
>>> possible modes a mapping can be used in.
>>>
>>>> I would instead suggest a function mappings_dump()/mappings_add():
>>>> first is similar to `nvim[_buf]_get_keymap` and should dump all
>>>> mappings as a list of maparg()-like dictionaries. Second should define
>>>> mappings being given a list of them. Of course, this means that
>>>> dictionaries need to be fixed to allow correctly saving/restoring.
>>>>
>>>> The advantages:
>>>>
>>>> 1. Easier to implement. Code for creating a maparg() dictionary is
>>>> already there, iterating over all mappings is not a problem. Results
>>>> needs to be incompatible with maparg() or use additional keys though:
>>>> e.g. Neovim altered the contents of `noremap` and `buffer` keys: first
>>>> is now 0, 1 or 2 (you can't correctly restore a mapping if you can't
>>>> distinguish `map <script>` and `noremap`) and second is a buffer
>>>> number or zero.
>>>> 2. More flexible: you can save and restore everything, push or pop
>>>> individual mappings, create a temporary mapping which is just like
>>>> mapping X, but has `<Plug>(Translit3TemporaryMap)` lhs instead (to be
>>>> returned from `<expr>` mappings in order to select either plugin
>>>> behaviour or fall back to previously present user mapping instead).
>>>>
>>>> I can imagine other usages enable/disable or push/pop could not
>>>> achieve: generating configuration with mappings like :mkvimrc, but
>>>> allows doing adjustments (parsing `:mkvimrc` output is not fun,
>>>> especially if you want to be forward compatible), creating a plugin
>>>> which analyses how often different mappings are used (need to copy all
>>>> mappings to temporary then replace existing mappings with plugin
>>>> ones).
>>>> 3. This is also forward compatible: just need to state in the
>>>> documentation that new significant keys may be added in the future to
>>>> the dictionaries so they should be preserved.
>>>
>>> I don't see much use for this. I can't think of a practical example how
>>> a plugin manipulates mappings it didn't create itself or even knows what
>>> they are for.
>>
>> Still Vim has :mkvimrc which does manipulate (dump) mappings from
>> third-party plugins. Also I need this functionality for some <expr>
>> mappings: if some condition is true (e.g. `>` is preceded by `-` (C,
>> completion) or transliteration mode was enabled, or transliteration
>> mode is enabled *and* character that does not start a new
>> transliteration sequence is a continuation of previous one) use plugin
>> mapping. If it is false, fall back to whatever was there previously,
>> including falling back to whatever mapping was there previously.
>>
>> Also check https://github.com/neovim/neovim/issues/6123, this is the
>> issue backing Neovim nvim_get_keymap() API function.
>>
>>>
>>> Another complication is that mappings can be added/removed by other
>>> mappings and by autocommands.
>>
>> I do not see how this complication is relevant to the discussion. I.e.
>> I do not see how this complication should affect usage or
>> implementation of both proposed changes.
>>
>>>
>>> Disabling and re-enabling mappings is definitely more efficient than
>>> removing and adding-back mappings.
>>
>> And it is also definitely both harder to implement and less flexible.
>
> Harder to implement, perhaps, but not necessarily less
> flexible. Though the discussion thus far has centered mostly
> on enable/disable functionality, there's nothing about the map
> handle interface that limits it to this. It could support
> query and execute functions, for instance. For cases in which
> you wish to keep the original behavior, but need to "wrap" it
> somehow, you could use the map handle to attach prolog/epilog
> callback functions to a map. Presumably, such callback
> functions (which could be either lambdas or funcrefs) would
> accept an argument that allowed them to obtain information
> about the original map, possibly even its exact lhs and rhs.
> The prolog callback would be even more useful if Vim provided
> a way (e.g., nonzero return) for it to abort the original map.

Enable, disable, query, execute plus two callbacks. *Four* functions
and two callbacks in place of just two simple functions, mostly using
the functionality that is already there. Five if you remember about
:mkvimrc and that somebody may want to replace that on top of new API:
query will need a mirror function for creating a mapping then.

This is going to introduce a big amount of bugs just to add the
flexibility which is naturally available through a much simpler
approach. Emulating everything you mention on top of current VimL
state plus mappings_dump()/mappings_load() / (mappings_clear()*) is
not going to make plugins considerably slower (as long as you can
operate on lists and use `map()`/`filter()`/etc: main VimL
optimization principle is "the less Ex commands the faster the code")
and I do not see any other benefits, except for "with some handles
implementation it may be slightly easier to pinpoint third-party
plugins' bugs".

* Found an issue in my proposal: `:execute 'unmap'` would not be easy
or efficient to use, so additionally need either `mappings_clear({list
to clear})` or make `mappings_load()` unmap mappings when rhs key is
missing.

>
> Hmm... This may be overkill, but it might even be possible to
> support the idea of a "virtual map handle": i.e., a handle not
> to a specific map, but to a *set* of maps matching certain
> criteria: e.g., <buffer>, <expr>, maps matching a mode mask,
> maps starting with specific char(s), etc... Once such a
> virtual handle had been obtained, a single call would suffice
> to enable/disable, or even attach callbacks to all maps in the
> set. Of course, some operations (e.g., execute) would be
> permitted only on non-virtual (single map) handles.

And this is just mappings_dump() + filter() with my approach without
any need to invent a new DSL to describe criterias (or not invent DSL,
but use VimL expressions which would be just as efficient as
filter()). If I got it right then plus some way to attach callbacks to
"new mapping defined" event to keep "callback attached" state.

>
> Sincerely,
> Brett Stahlman
>
>>
>>>
>>> --
>>> BLACK KNIGHT: I'm invincible!
>>> ARTHUR: You're a looney.
>>> "Monty Python and the Holy Grail" PYTHON (MONTY) PICTURES LTD
>>>
>>> /// Bram Moolenaar -- Bram@Moolenaar.net -- http://www.Moolenaar.net \\\
>>> /// sponsor Vim, vote for features -- http://www.Vim.org/sponsor/ \\\
>>> \\\ an exciting new programming language -- http://www.Zimbu.org ///
>>> \\\ help me help AIDS victims -- http://ICCF-Holland.org ///

--
--
You received this message from the "vim_use" maillist.
Do not top-post! Type your reply below the text you are replying to.
For more information, visit http://www.vim.org/maillist.php

---
You received this message because you are subscribed to the Google Groups "vim_use" group.
To unsubscribe from this group and stop receiving emails from it, send an email to vim_use+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.