[PATCH v2] msvcrt: Implement strtod without using 'long double'.

[Erich E. Hoover]

v2: Fix test results on 32-bit Debian *sigh*

Fixes the "regression" with cl.exe caused by
0e183cc3c0d3b6f89f79047cdd71c389afc75073.
( https://bugs.winehq.org/show_bug.cgi?id=48160 )

Best,
Erich

[Alexandre Julliard]

"Erich E. Hoover"  writes:
> v2: Fix test results on 32-bit Debian *sigh*
>
> Fixes the "regression" with cl.exe caused by
> 0e183cc3c0d3b6f89f79047cdd71c389afc75073.
> ( https://bugs.winehq.org/show_bug.cgi?id=48160 )
>
> Best,
> Erich
>
> From 9742794ff106d76713f80c6497accc1fc78ae6c9 Mon Sep 17 00:00:00 2001
> From: "Erich E. Hoover" 
> Date: Wed, 18 Dec 2019 15:48:11 -0700
> Subject: msvcrt: Implement strtod without using 'long double'.
>
> Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=48160
> Signed-off-by: Erich E. Hoover
>
> dlls/msvcrt/string.c       | 60 ++++++++++++++++++++++++++------------
>  dlls/msvcrt/tests/string.c | 44 ++++++++++++++++++++++++++--
>  2 files changed, 83 insertions(+), 21 deletions(-)

Thanks! Do we still need the control87 call then?

Also we should do the same changes in wcstod().

[Erich E. Hoover]

On Thu, Dec 19, 2019 at 2:58 AM Alexandre Julliard  wrote:
> "Erich E. Hoover"  writes:
> . ..
> > Subject: msvcrt: Implement strtod without using 'long double'.
> >
> > Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=48160
> > Signed-off-by: Erich E. Hoover
>
> ...
>
> Thanks! Do we still need the control87 call then?

No problem!  Yes, unfortunately calculating 1.7976931348623158e+308
(DBL_MAX) requires 64-bit doubles*.  This works just fine on x86-64,
but x86-32 defaults to 53-bit precision.  However, it would be
possible to only change the precision if we get into the range of
values where it becomes a problem, though that might be a little
complicated to do properly.
> Also we should do the same changes in wcstod().

Would you like that as a separate patch or part of this one?

Best,
Erich

* I am unaware of a way to multiply 1.7976931348623158 by 1e+308 with
53-bit doubles.

[Alexandre Julliard]

"Erich E. Hoover"  writes:
> On Thu, Dec 19, 2019 at 2:58 AM Alexandre Julliard  wrote:
> > "Erich E. Hoover"  writes:
> > . ..
> > > Subject: msvcrt: Implement strtod without using 'long double'.
> > >
> > > Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=48160
> > > Signed-off-by: Erich E. Hoover
> >
> > ...
> >
> > Thanks! Do we still need the control87 call then?
>
> No problem!  Yes, unfortunately calculating 1.7976931348623158e+308
> (DBL_MAX) requires 64-bit doubles*.  This works just fine on x86-64,
> but x86-32 defaults to 53-bit precision.  However, it would be
> possible to only change the precision if we get into the range of
> values where it becomes a problem, though that might be a little
> complicated to do properly.
> > Also we should do the same changes in wcstod().
>
> Would you like that as a separate patch or part of this one?

A separate patch would be nice. I've committed the first one with a few
changes.