Loading ...
Sorry, an error occurred while loading the content.
 

Re: if_lua patch

Expand Messages
  • ZyX ZyX
    ... Pulling patches from the repository and saving them there is convenient. When I was adding third argument to sort() patch was also not big but I still
    Message 1 of 13 , Sep 1, 2012
      2012/9/1 Luis Carvalho <lexcarvalho@...>:
      >> Wondering whether you have repository with all these changes.
      >
      > No, since the patches are few and far between.

      Pulling patches from the repository and saving them there is
      convenient. When I was adding third argument to sort() patch was also
      not big but I still cloned vim to bitbucket. This decision appeared to
      be right when I started to extend python interface.

      Even if you don’t want to create commits, do use “hg diff”. Two
      patches for two files with requirement to switch to another directory
      before applying each one is very inconvenient.

      >> After encountering this error in if_py* found it in your interface also:
      >>
      >> vim -u NONE -c 'let d={} | lua vim.eval("d")[""]=1' -c 'echo d'
      >>
      >> outputs “{'': 1.0}” while it should output error and “{}”: vim does not allow empty keys in dictionaries (it does not look like it breaks something except that doing “d['']” results in an error).
      >
      > I don't think that's an error in if_lua: the documentation doesn't say
      > anything about dictionaries not allowing empty strings as keys -- :help
      > Dictionary only says that 'a key is always a String' -- and hence the absence
      > of error when printing d in your example. What is not allowed is for you to
      > read or set empty keys in Vim (using 'let'), but I don't want to replicate
      > that in if_lua. For instance, you can do this:
      >
      > vim -u NONE -c 'let d={} | lua vim.eval("d")[""] = "empty"'
      > -c 'for [k,v] in items(d) | echo "|".k."| -> ".v | endfor'
      >
      > So, to be consistent, if you set an empty key in if_lua, you can read or reset
      > it there.

      Yes, you can do “for [k, v] in items(d)”. But you can’t do “for k in
      keys(d) | let v=d[k]”. I can’t prove that this won’t break anything,
      maybe you can?

      Help does not mention this probably because error message is clear
      enough, I constantly see such things in a doc: “:h E???” exists, but
      in the text there is nothing concerning error message.

      >
      >> And you have missed another error:
      >>
      >> vim -u NONE -c 'lua vim.eval("g:")["input"]=vim.funcref("tr")' -c 'call input("Yes?")'
      >>
      >> results in an error “not enough arguments to tr” while it should result in an error “funcref variables must start with a capital” (see my patch to python interface [1] and extend() function [2]).
      >
      > There's no error here either. The problem is that tr() expects three
      > arguments, and not that the variable must start with a capital letter (g:input
      > already exists.) Try this:
      >
      > vim -u NONE -c 'lua vim.eval("g:").input=vim.funcref("tolower")'
      > -c 'echo input("Yes?")'

      It is just an example. Vim has very weird way of disallowing
      overriding built-in and user functions: a check everywhere you can add
      a value to a dictionary. The fact that you *allow overriding built-in
      functions* is an error. Not the fact that “tr” expects three arguments
      or something else.

      Existence of your behavior in lua interface and my in python (unlike
      my patch to extend() patch to if_py* was not merged) defeats the
      purpose of such checks.

      >
      > I think you're mixing the fact that Vim Funcref variables should start with a
      > capital letter. Funcrefs in lua are just a value; for instance, this should
      > work:
      >
      > :lua print(vim.funcref"tr"("<tag>", "<>", "{}"))
      > " {tag}

      No, I am not. I do not care how you can name funcrefs in lua, you must
      not allow using lua to override built-in or user functions by adding
      something to scope dictionary. The proper way of fixing this all is
      modifying parser, but modifying lua interface is much simpler. If you
      can modify the parser go ahead and get rid of E704 and E705: in the
      current state there is exactly no way to safely assign funcref to a
      variable, only to dictionary key/list item. I can see why parser does
      this (deref_func_name: it first tries to get a funcref variable and
      returns name as-is in case of failure), but unsure how to fix it
      properly (quick fix is returning name as-is if find_internal_func
      returned something other than -1 or find_func — other than NULL). I
      also don’t think this fix will be accepted by Bram: it only pretends
      to solve a problem as now masking functions by variables is replaced
      by masking variables by a functions:

      let input="abc"
      may work properly but

      function Input()
      return getchar()
      endfunction
      let Input=function('input')
      call Input("abc")
      won’t.

      --
      You received this message from the "vim_dev" 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
    • ZyX
      ... No, I meant exactly what I typed. Before reading docs expected behavior was either converting it to “{ 1 : abc }” or an error. -- You received this
      Message 2 of 13 , Sep 1, 2012
        > Right. I think you meant:
        >
        > :lua vim.eval('g:').list = vim.list{'abc'}
        >
        > or
        >
        > :lua vim.eval('g:').dict = vim.dict{key = 'abc'}
        >
        > since you forgot the key.

        No, I meant exactly what I typed. Before reading docs expected behavior was either converting it to “{"1": "abc"}” or an error.

        --
        You received this message from the "vim_dev" 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
      • Luis Carvalho
        ... OK, thanks for the suggestion. I m attaching a patch generated by hg diff . To summarize the main changes: - New funcref type, including new vim.funcref()
        Message 3 of 13 , Sep 1, 2012
          ZyX wrote:
          > > No, since the patches are few and far between.
          >
          > Pulling patches from the repository and saving them there is
          > convenient. When I was adding third argument to sort() patch was also
          > not big but I still cloned vim to bitbucket. This decision appeared to
          > be right when I started to extend python interface.
          >
          > Even if you don’t want to create commits, do use “hg diff”. Two
          > patches for two files with requirement to switch to another directory
          > before applying each one is very inconvenient.

          OK, thanks for the suggestion. I'm attaching a patch generated by 'hg diff'.
          To summarize the main changes:

          - New funcref type, including new vim.funcref() function
          - vim.dict() and vim.list() now accept an initializer table as argument
          - Updated docs
          - A few bug fixes (following this thread)


          > > I don't think that's an error in if_lua: the documentation doesn't say
          > > anything about dictionaries not allowing empty strings as keys -- :help
          > > Dictionary only says that 'a key is always a String' -- and hence the absence
          > > of error when printing d in your example. What is not allowed is for you to
          > > read or set empty keys in Vim (using 'let'), but I don't want to replicate
          > > that in if_lua. For instance, you can do this:
          > >
          > > vim -u NONE -c 'let d={} | lua vim.eval("d")[""] = "empty"'
          > > -c 'for [k,v] in items(d) | echo "|".k."| -> ".v | endfor'
          > >
          > > So, to be consistent, if you set an empty key in if_lua, you can read or reset
          > > it there.
          >
          > Yes, you can do “for [k, v] in items(d)”. But you can’t do “for k in
          > keys(d) | let v=d[k]”. I can’t prove that this won’t break anything,
          > maybe you can?
          >
          > Help does not mention this probably because error message is clear
          > enough, I constantly see such things in a doc: “:h E???” exists, but
          > in the text there is nothing concerning error message.

          I agree that there must be a reason for the parser to avoid empty strings, but
          then the documentation should change to reflect this behavior. Ideally this
          behavior should be coded in the API, probably making dict_add check the
          dictitem argument. While I still maintain that that's not an if_lua issue,
          I've updated the code to check for empty keys when setting dict entries, just
          in case it breaks something and for consistency.

          > >> And you have missed another error:
          > >>
          > >> vim -u NONE -c 'lua vim.eval("g:")["input"]=vim.funcref("tr")' -c 'call input("Yes?")'
          > >>
          > >> results in an error “not enough arguments to tr” while it should result in an error “funcref variables must start with a capital” (see my patch to python interface [1] and extend() function [2]).
          > >
          > > There's no error here either. The problem is that tr() expects three
          > > arguments, and not that the variable must start with a capital letter (g:input
          > > already exists.) Try this:
          > >
          > > vim -u NONE -c 'lua vim.eval("g:").input=vim.funcref("tolower")'
          > > -c 'echo input("Yes?")'
          >
          > It is just an example. Vim has very weird way of disallowing
          > overriding built-in and user functions: a check everywhere you can add
          > a value to a dictionary. The fact that you *allow overriding built-in
          > functions* is an error. Not the fact that “tr” expects three arguments
          > or something else.
          >
          > Existence of your behavior in lua interface and my in python (unlike
          > my patch to extend() patch to if_py* was not merged) defeats the
          > purpose of such checks.

          Again, the issue here is not that funcref variables must start with a capital
          letter -- only the parser should check that, not if_lua -- but that if_lua is
          assigning funcrefs to a "builtin" scope dictionary (g: or l:). The fix is
          straightforward: if_lua now checks if dict->dv_scope == VAR_DEF_SCOPE and if
          the value is a funcref and complains if that's the case.

          > > I think you're mixing the fact that Vim Funcref variables should start with a
          > > capital letter. Funcrefs in lua are just a value; for instance, this should
          > > work:
          > >
          > > :lua print(vim.funcref"tr"("<tag>", "<>", "{}"))
          > > " {tag}
          >
          > No, I am not. I do not care how you can name funcrefs in lua, you must
          > not allow using lua to override built-in or user functions by adding
          > something to scope dictionary. The proper way of fixing this all is
          > modifying parser, but modifying lua interface is much simpler. If you

          As I said, you shouldn't really care about Lua "naming" funcrefs, since that's
          a moot point (funcrefs in Lua are just values, they have no names.) But you're
          right that the issue is checking for the dictionary to be a scope; see comment
          above.

          > can modify the parser go ahead and get rid of E704 and E705: in the
          > current state there is exactly no way to safely assign funcref to a
          > variable, only to dictionary key/list item. I can see why parser does
          > this (deref_func_name: it first tries to get a funcref variable and
          > returns name as-is in case of failure), but unsure how to fix it
          > properly (quick fix is returning name as-is if find_internal_func
          > returned something other than -1 or find_func — other than NULL). I
          > also don’t think this fix will be accepted by Bram: it only pretends
          > to solve a problem as now masking functions by variables is replaced
          > by masking variables by a functions:
          >
          > let input="abc"
          > may work properly but
          >
          > function Input()
          > return getchar()
          > endfunction
          > let Input=function('input')
          > call Input("abc")
          > won’t.

          You lost me in this last example and the masking comment, but I believe I do
          not need to change the parser (and shouldn't need to.)

          ZyX wrote:
          > > Right. I think you meant:
          > >
          > > :lua vim.eval('g:').list = vim.list{'abc'}
          > >
          > > or
          > >
          > > :lua vim.eval('g:').dict = vim.dict{key = 'abc'}
          > >
          > > since you forgot the key.
          >
          > No, I meant exactly what I typed. Before reading docs expected behavior was either converting it to “{"1": "abc"}” or an error.

          Ah, I see: you wanted to check for an error since {'abc'} is not valid in Vim.
          I've changed the behavior to be more in line with Lua (and Vim) since numbers
          are usually converted to strings: now vim.dict{'abc'} returns {'1': 'abc'}.

          Thanks once again for your feedback and thorough tests!

          Cheers,
          Luis

          --
          Computers are useless. They can only give you answers.
          -- Pablo Picasso

          --
          Luis Carvalho (Kozure)
          lua -e 'print((("lexcarvalho@..."):gsub("(%u+%.)","")))'

          --
          You received this message from the "vim_dev" 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
        Your message has been successfully submitted and would be delivered to recipients shortly.