Re: Bug in pcall? or not?

[David Jones <drj@...>]

On 12 Oct 2006, at 07:33, Nick Gammon wrote:

> Is this a bug?

No.

> --------------
> mt = {}
> mt.__index = mt
> mt.test = function (a, b, c)
>           print "test"
>           end  -- function mt.test
>
> t = {}
> setmetatable (t, mt)
>
> ok, result = pcall (t:test)  --> error: function arguments expected  
> near `)'
> --------------

The a:b syntax isn't valid as an expression, only as part of a  
function call.  That means that (a:b)(x, y, z) can't be explained by  
considering (a:b) (production of a function value) and (x, y, z)  
(calling a function with specified args) separately.  This is a shame  
in my opinion.

Years ago I suggested that a:b be sugar for

function (...)
  return a.b(a, ...)
end

Of course, we couldn't use '...' like that in those days.  And also  
of course I'm not suggesting that a new function closure be created  
for the common case of a:b(x,y,z).

This would mean then that a:b would be a valid expression.  I think  
your pcall case provides good motivation for this.

(I think it's a bit trickier than my simple rewrite suggests, as 'a'  
should probably be an upvalue of the created closure).

drj