Fix signed/unsigned mismatch warning on gcc.

[Lastique]

No description provided.

[djowel]

Why did you add the logic with MaxDigits < 0 ? I don' think this will ever happen in this code branch.

[Lastique]

On Tue, Jul 14, 2015 at 12:50 PM, Joel de Guzman notifications@github.com
wrote:

In include/boost/spirit/home/qi/numeric/detail/numeric_utils.hpp
#140 (comment):

@@ -316,7 +316,7 @@ namespace boost { namespace spirit { namespace qi { namespace detail
if (!Accumulate)
{
// skip leading zeros

•            while (it != last && *it == '0' && leading_zeros < MaxDigits)
  

•            while (it != last && *it == '0' && (MaxDigits < 0 || leading_zeros < static_cast< std::size_t >(MaxDigits)))
  

Why did you add the logic with MaxDigits < 0 ? I don' think this will ever
happen in this code branch.

1. I didn't find any restrictions on the MaxDigits values. From the
   specialization below it seems MaxDigits == -1 is allowed and means "no
   limit". I don't know what behavior should be for other negative values, so
   I extended the same semantics on them as well. Let me know if this
   assumption is wrong.
2. I found only one specialization of extract_int, which is the one below -
   for MinDigits == 1 and MaxDigits == -1. This specialization won't be used
   for e.g. MinDigits == 3 && MaxDigits == -1 which means the generic template
   will be used and the 'while' condition should be prepared for negative
   values.

[djowel]

Indeed. You are right. I'll review the code some more. It's been a while.

[Lastique]

Ping? Would be great if this fix made it to 1.59.

[djowel]

I'm not comfortable with the code change and possible behaviour change, unless there are some tests that exploit the new code, if it is indeed a bug without it. In that case, then perhaps it's best to have that in another PR.

[Lastique]

The original code was comparing size_t with an int. The negative int would be converted to a large unsigned number in this case, so the behavior was "to skip up to 4 billion something leading zeros". The current behavior is "to skip all leading zeros", which is the same for all practical purposes.

I can change this PR to explicitly cast the int to size_t, which will remove the warning and keep the 4 billion quirk. It just doesn't make much sense to me. Should I do it?

[djowel]

OK, I studied the code a bit more. Pardon my being slow. MaxDigits < 0 is indeed a good idea and that branch will be optimized away by the compiler quite nicely (being const).

[djowel]

Thanks, Andrey.