numeric_limits

[Gabriel Dos Reis]

Richard Henderson <rth@twiddle.net> writes:

| I think I'm nearly done with the cleanup.
| 
| There are three floating point target hooks left:
| 
|   __glibcpp_float_has_denorm_loss
|   __glibcpp_float_tinyness_before
| 	I suspect these can be defined to has_denorm == denorm_present.
| 	I need to survey to see how correct this is.

Agreed.

|   __glibcpp_float_traps
| 	This is true for the overwhelming majority of hardware
| 	implementations, but is false for most software implementations.
| 	I guess there's nothing to do but leave it as-is.

Agreed.

| Additionally, the config subdirectory could stand to be cleaned
| up to remove definitions that are no longer relevant.  In particular,
| 
| 	__glibcpp_long_bits
| 	__glibcpp_long_double_bits
| 	__glibcpp_wchar_t_bits
| 
| In most cases, this would completely empty the cpu_limits.h file.
| Is it preferred to remove the empty file, or leave the configury
| alone for the nonce?

I would vote for removing the empty file.  The hook
__glibcpp_integer_traps may be overriden later in such a file if a
target had a need (which, if I understood you, isn't currently the case).

-- Gaby