Re: Feature request - attribute type should not be altered for terminal rules

On 5/10/2012 3:00 PM, Vitaly Budovski wrote:
> Hi all,
> 
> I've discussed this with Joel on IRC, but am posting this to the list
> so others can comment.
> Currently code such as this results in unexpected behaviour (for me).
> 
> #define BOOST_SPIRIT_ACTIONS_ALLOW_ATTR_COMPAT
> 
> int result;
> parse(first, last, word[_val = if_else_(_1 > 1000u, -1, 5)], result);
> 
> This results in a warning about comparison between signed and unsigned
> types (_1 > 1000u).
> This is because the attribute type is converted to int to match the
> type of the result variable, even though the type returned by word is
> unsigned.
> 
> I would like to request a change in the behaviour for terminal rules,
> if possible, so that this no-longer occurs. The type of _1 would now
> be unsigned, and the comparison would happen without warnings. The two
> possible branches output a type that is compatible with the type of
> the output variable and so it succeeds.
> 
> The code below, however, should result in a warning or error:
> 
> int result;
> parse(first, last, dword, result); // Difference in sign
> 
> or
> 
> unsigned result;
> parse(first, last, qword, result); // Might overflow
> 
> Is this feasible?

I probably did not communicate well on IRC. Let me try again.
So, what is the attribute of the int_ parser? By default, it is
int. However, if the user passes in his own attribute, the int_
parser uses its type and it becomes its attribute. Basically, for
any parser P, we have:

  1) DA: the default attribute type of P (for int_ it is int)
  2) AA: The actual attribute of P (the one that the user actually provides)

Essentially, P is not tied to its DD (default attribute type). For as long as
AA (The actual attribute) is 'compatible with' the parser P. It will be
accepted. For example, your AA can be short for parser int_. That is OK.
int_ will try to parse as much as it can, but report an error when the
short overflows. For that matter, it should also be possible to pass in
user defined types such as bignum for int_. When passed a bignum, the int_
parser will use it, parse the input, and compute the result (compile the
bignum).

Semantic actions use the AA attribute when you #define
BOOST_SPIRIT_ACTIONS_ALLOW_ATTR_COMPAT. That way, if you pass in a
bignum, your SA will get the bignum that int_ parsed for you.

(In your case, you are getting the warning because you passed in
an int which the word parser takes in as its AA and passes that to
your SA as _1 and you compare to an unsigned)

Now, imagine if we force int_ to always pass _1 with type int.
If you passed a bignum, that will be lossy!

  1) the int_ parser parses a big number, say
     "1234567890123456789012345678901234567"
  2) compiles that into a bignum
  3) converts it to an int (oops)
  4) passes it to your SA

Step no. 3 is problematic. The bignum will be downsized to an int.
Step no. 4 is also problematic because now, there is no way to get
the actual bignum result that was parsed.

(Aside: in spirit-3, BOOST_SPIRIT_ACTIONS_ALLOW_ATTR_COMPAT will
be the default behavior).

Regards,
-- 
Joel de Guzman
http://www.boostpro.com
http://boost-spirit.com




------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/