On 2018-08-10 14:43, Houder wrote:
On Fri, 10 Aug 2018 13:42:54, Corinna Vinschen wrote:
[snip]
> Note: the following line must be changed in STC-FENV.c (the STC that was
> attached to the bug report -- the last one above).
>
> from:
> const int xxx =3D 0x3d; // FE_ALL_EXCEPT on Linux, i.e. the denormal-...
> // exception is excluded on Linux
>
> to:
> const int xxx =3D 0x3f; // Cygwin allows the denormal-operand exception;
> // Linux (Andreas Jaeger) does not.
> // This exception is not defined by the IEEE 754
> // standard (Floating-Point Arithmetic)
Shall we change that? A patch would be nice. ;)
That is possible ...
As on Linux, the denormal-operand exception will always be masked
after this
change ... (both in the default env. and in the "nomask" env.)
Meaning that, after this change, this exception can * no longer * be
enabled
through the API of fenv.h
Objections? Anyone?
The modication would require changing:
winsup/cygwin/fenv.cc (_feinitialise() )
winsup/cygwin/include/fenv.h (FE_ALL_EXCEPT)
GRRR! The file encoding of fenv.h is "cp1252" because of 2 characters
in this
line:
Intel® 64 and IA-32 Architectures Software Developer’s Manuals:
... part of a comment at the beginning of the file.
(the registered trademark sign (u00ae) is encoded as 0xae (cp1252),
while it
would be: 0xc2 0xae, in utf-8,
the right single quotation mark (u2019) is encoded as 0x92 (cp1252),
but in
utf-8 it would be: 0xc2 0x80 0x98)