Another rant

[Xen]

Colin Law schreef op 17-11-2017 18:36:
> On 17 November 2017 at 15:45, Xen <list at xenhideout.nl> wrote:
>> Liam Proven schreef op 17-11-2017 0:18:
>> 
>>> Agreed with all of that. It's caused a whole world of problems, too.
>> 
>> 
>> I just lost an entire day to C because
>> 
>> a) I'm a novice
>> b) Older versions of GCC had different behaviour
> 
> Was that a change to the c/c++ standard rather than a gcc issue?  If
> so then you would have had the same issues on Windows.

No apparently the C standard leaves it as undefined but GCC chooses to 
implement it in a certain way OR it is even platform (CPU) specific.

GCC used to allow -1 << 4 but a patch in 2015 now ensures errors when 
doing so.

One must now use ~0 << 4 to achieve the same effect if that even works 
because I think it doesn't
and you have to use ~0U or something.

I think it even treats 0 as signed, ~0 as -1, and then fails.

But I am confused, << is apparently undefined and >> is defined as sign 
extension.

But honestly this only goes to show the quagmire this language is.

My program error probably resulted from me depending on >>

and seeing different behaviour across 2 platforms.

GCC now clearly defines it as arithmetic shift right.

"If E1 has a signed type and a negative value, the resulting value is 
implementation-defined."  (C89 standard)

It clearly also didn't change in C90.

It also did not change in C11.

https://port70.net/~nsz/c/c11/n1570.html

So no, your allusion is incorrect




Yet the C standard does say that the implementation can be 
platform-specific.