Catch div 0

[Johannes Loher]

On Sunday, 21 July 2019 at 01:59:53 UTC, mogu wrote:
> On Saturday, 20 July 2019 at 21:11:34 UTC, Johannes Loher wrote:
>> Am 20.07.19 um 19:31 schrieb mogu:
>>> Should d make div 0 a runtime exception? Or as an optional 
>>> feature? I never want my program crashes only beause of 
>>> reading a wrong number from runtime configurations.
>>> 
>>
>> I suggest you take a look at std.experimental.checkedint 
>> (https://dlang.org/phobos/std_experimental_checkedint.html). 
>> It allows you to specify what exact behavior you want to have 
>> on a very fine granular level and also defines a few very 
>> useful presets, e.g. `Throw`, which looks very much like you 
>> want:
>>
>> ```
>> import std.exception : assertThrown;
>> auto x = -1.checked!Throw;
>> assertThrown(x / 0);
>> ```
>
> Thanks. But if I use checked int, I must let all my libs 
> support checked int. Otherwise I must take care of all the 
> parts using ints from checked in order to safely interacting 
> with other libs.
> As a designer, I wish other logic programmers will not break 
> down the whole program suddenly. But now I have to use 
> multi-process instead of only a single one to handle it.

I do not think the situation is as bad as you describe it. In 
particular the interaction of checked int with regular ints works 
reasonably well.

But yes, you are right, code written by others can still result 
in the whole program being stopped by SIGFPE.

For better or worse, checked int is the best solution you have. 
Throwing on division by 0 for regular ints will never get into D 
because that would mean that regular integer division (and modulo 
operation) is not nothrow anymore, which would prevent almost any 
code from being nothrow, rendering nothrow totally useless. Also 
it would make division slower because it would need to be checked 
in the language while currently the SIGFPE is basically raised by 
the processor.