__DBL_HAS_DENORM__ et al required in recent libstdc++

__DBL_HAS_DENORM__ et al required in recent libstdc++

[Tim Prince]

Builds of gcc-4.3 libstdc++ on cygwin have begun failing with these 
undefined symbols for FLT,DBL,LDBL in std_limits.h.  I don't find any 
explanation of it.  Is it a particular problem for cygwin, given that 
the include files for LDBL are incomplete, and don't distinguish between 
DBL and LDBL?  Are these values intended to reflect options such as 
-ffast-math and -mfpmath which would affect use of subnormals (formerly 
called denorm)?

Re: __DBL_HAS_DENORM__ et al required in recent libstdc++

[Gabriel Dos Reis]

Tim Prince <timothyprince@sbcglobal.net> writes:

| Builds of gcc-4.3 libstdc++ on cygwin have begun failing with these
| undefined symbols for FLT,DBL,LDBL in std_limits.h.  I don't find any
| explanation of it.

They are built-in CPP macros.

-- Gaby

Re: __DBL_HAS_DENORM__ et al required in recent libstdc++

[Tim Prince]

Gabriel Dos Reis wrote:
> Tim Prince <timothyprince@sbcglobal.net> writes:
> 
> | Builds of gcc-4.3 libstdc++ on cygwin have begun failing with these
> | undefined symbols for FLT,DBL,LDBL in std_limits.h.  I don't find any
> | explanation of it.
> 
> They are built-in CPP macros.
> 
> -- Gaby
> 
> 

Thanks, but I'm not certain what that means.  I would have taken your 
comment to mean that they would be included among the pre-defined 
macros. I see the function builtin_define_float_constants()  and I see 
the file modified date Nov 1, but I don't see any ChangeLog references 
in recent months indicating what was changed then, why these macros 
would suddenly become undefined a few weeks later, or suddenly become 
required by libstdc++, if always undefined.  Not even any documentation 
which would confirm my guess that
-D__LDBL_HAS_DENORM__=0  //(setting apparently immaterial)
-D__DBL_HAS_DENORM__=1
-D__FLT_HAS_DENORM__=1
would be the expected settings.  Are those supposed to be added to 
CXXFLAGS?  Why not documented, why not done automatically?
Adding these to CXXFLAGS does avoid the complaints about those macros, 
but immediately after, there is an error thrown about illegal 
overloading of std::num_get in libstdc++-v3/include/bits/locale_facets.h

Re: __DBL_HAS_DENORM__ et al required in recent libstdc++

[Gabriel Dos Reis]

Tim Prince <timothyprince@sbcglobal.net> writes:


[...]

| Thanks, but I'm not certain what that means.  I would have taken your
| comment to mean that they would be included among the pre-defined
| macros. I see the function builtin_define_float_constants()  and I see
| the file modified date Nov 1, but I don't see any ChangeLog references

grep is your friend.

2005-12-13  Carlos O'Donell <carlos@codesourcery.com>

        * c-cppbuiltin.c (builtin_define_float_constants): Add
        fp_cast parameter, pass to builtin_define_with_hex_fp_value.
        Define __FLT_HAS_DENORM__, __DBL_HAS_DENORM__,
        __LDBL_HAS_DENORM__.
        (builtin_define_with_hex_fp_value): Use fp_cast when building
        macro.
        (c_cpp_builtins): If flag_single_precision_constant then set
        fp_cast
        to "((double)%sL)" otherwise "%s".

-- Gaby

Re: __DBL_HAS_DENORM__ et al required in recent libstdc++

[Gabriel Dos Reis]

Tim Prince <timothyprince@sbcglobal.net> writes:

[...]

Would you be more polite the next time and refrain from setting the
"Reply-To" field to an invalid address when you ask for assistance on
GCC list?  Thanks, 

-- Gaby