Re: inconsistencies between arithmetic and boolean operators

[Adrien de Croy <adrien@...>]

I guess I'm trying to shield my users from having to consider type.

Lua already does a certain amount of this as described in the section of 
the lua documentation on lua types.  E.g it automatically converts any 
type to a number for arithmetic operations, and types of variables are 
mutable.

I agree it's perfectly valid to compare strings, although the numeric 
representation of strings can have some unexpected consequences in a 
string compare, since it also takes into account length of string, and 
would make a mess of signs.

remember that all this is in the context of a __cast metamethod.  the 
concept of a __cast operator is to allow automatic conversion of types 
where this is explicitly provided for by the host by registering the 
metamethod.  If you don't register the __cast metamethod, you don't get 
automatic type conversion.

I coded it up quickly, and it now allows me to compare objects as strings.

my problem is I have string classes in C++ with methods that I want to 
expose to lua, yet also have lua to be able to treat them as a string.  
E.g I publish my string object to lua as a table. this way I can still 
call members on it, but if I want to use it as a string at all, it needs 
to be converted into a lua string.

I wrote a __cast metamethod to hook the tostring and tonumber calls, and 
this then allowed me to pass my table-representing-a-string into 
functions that wanted a string, but other operators didn't work, such as 
equality, lt, gt etc.

I could register metamethods for these operators as well, but if we are 
thinking about registered automatic type conversion, that should be able 
to do it as well, and saves a lot of subsequent metamethod lookups.

otherwise if I went with strict lua syntax etc, I'd have to get my users 
to write lua code such as

function filter(Session)

   if(Session.UserBalance.AsNumber() < 
WinGate.Globals.MinBalance.AsNumber()) then
      return false;
   end;

   if(Session.Username.AsString() == "adrien")then
      return true;
   end

   if(string.find(Session.Request.Header["User-Agent"].AsString(), 
"gecko"))then
      return true;
   end

   -- block access  
   return false;
end

rather than

function filter(Session)

   if(Session.UserBalance < WinGate.Globals.MinBalance) then
      return false;
   end;

   if(Session.Username == "adrien")then
      return true;
   end

   if(string.find(Session.Request.Header["User-Agent"], "gecko"))then
      return true;
   end

   -- block access  
   return false;
end

I guess it's not so bad really, but I want to use a consistent syntax 
for object access from Lua as I do from my own expression evaluation 
code in the host application, which doesn't need casting because it 
knows whether the object is being indexed or not already.  The constant 
calls to AsString() and AsNumber() get a bit tedious.  If someone leaves 
them out, they get a policy bug.

It seems that lua has so much scope for overriding behaviour with 
metamethods, that something like this shouldn't be so far out of whack?

Adrien





Torsten Karwoth wrote:
> > boolean operators insist however that the 2 operands have the same type,
> > else the result is zero.  This is checked first.
> >
> > This means that the following code
> >
> > var1 = "3";
> > var2 = var1 * 2;
> >
> > is valid and var2 will have the numeric value of 6
> >
> > however
> >
> > var1 = "3"
> > if(var1 < 4) then
> > ...
> >
> > is invalid, since var1 and 4 have different types, the result is an
> > error.  If you wanted this to "work" you'd need to write
> >
> > var1 = "3"
> > if((var1 * 1) < 4) then
> > ...
> >     
>
> Thats not 100% correct ;-)
>
> first, you can also write
>   var1 = "3"
>   if tonumber(var1) < 4 then
>
> But, there is a difference between multiplication and compare...
> You can only multiply a numeric value, so it is IMHO ok to try to
> convert to a number 'automagically'.
> But you can also compare strings... so why dont you want to convert the
> second operand in your example to be converted to a string instead?
>
>   var1 = "3"
>   if (var1 < "4") then
>
> is absolute valid code.
>
> so, there cannot be an automatic type conversion IMHO.
>
> HTH,
>
>    Torsten
>   

--
Adrien de Croy - WinGate Proxy Server - http://www.wingate.com