• user001 (35/35) Feb 13 2016 There seems to be some invalid code generation for std.format for
• user001 (4/4) Feb 13 2016 Disregard all that above, looks like someone used a "double"
• jkpl (7/12) Feb 13 2016 yes the line you highlight is for DMD64 (windows of course),

user001 <youdontneedthis dont.com> writes:

There seems to be some invalid code generation for std.format for 
type of "real". Relavant links:

https://github.com/D-Programming-Language/phobos/blob/master/std/format.d#L1713
https://github.com/D-Programming-Language/phobos/blob/master/std/math.d#L5118


movabs rbx,7FF0000000000000                | <-- problem, should 
be 0x7FFF
and rax,rbx                                |
cmp rax,rbx                                |
jne test.7FF6FF4762E4                      |
mov rdx,qword ptr ss:[rbp-230]             |
movabs rcx,FFFFFFFFFFFFF                   |
test rdx,rcx                               |
je test.7FF6FF4762E4                       |
lea rax,qword ptr ds:[7FF6FF4C5C02]        | ;7FF6FF4C5C02:"nan"
mov qword ptr ss:[rbp-2C8],rax             |
mov qword ptr ss:[rbp-2D0],3               |
jmp test.7FF6FF476335                      |
mov rsi,qword ptr ss:[rbp-230]             |
movabs rax,7FFFFFFFFFFFFFFF                |
and rsi,rax                                |
cmp rsi,rbx                                |
jne test.7FF6FF476335                      |
fld tword ptr ss:[rbp-228]                 |
fldz                                       |
fcompi st1                                 |
fstp st0                                   |
ja test.7FF6FF47631B                       |
jp test.7FF6FF47631B                       |
lea rdx,qword ptr ds:[7FF6FF4C5C06]        | ;7FF6FF4C5C06:"inf"
mov ebx,3                                  |
jmp test.7FF6FF476327                      | ;format.d:1495
lea rdx,qword ptr ds:[7FF6FF4C5C0A]        | ;7FF6FF4C5C0A:"-inf"

Seems to compile fine with DMD at least with simpler code with 
just an if statement. Also this issue doesn't exist for 32-bit 
x86.

user001 <youdontneedthis dont.com> writes:

Disregard all that above, looks like someone used a "double" 
which uses 0x7FF0 to check if infinite. Not sure why the code 
runs fine on x86 though, different library maybe?

https://github.com/D-Programming-Language/phobos/blob/master/std/format.d#L1709

jkpl <jkpl nowhere.fr> writes:

On Saturday, 13 February 2016 at 20:50:22 UTC, user001 wrote:
 Disregard all that above, looks like someone used a "double" 
 which uses 0x7FF0 to check if infinite. Not sure why the code 
 runs fine on x86 though, different library maybe?

 https://github.com/D-Programming-Language/phobos/blob/master/std/format.d#L1709

yes the line you highlight is for DMD64 (windows of course), 
which explains well why x86 is OK since it uses the C runtime 
library by DigitalMars (aka snn.lib). Now I'm not quite sure if 
it should be

T tval = val;

because of the comment just after:

 convert early to get "inf" in case of overflow