My Kingdom For ...

[Don Clugston]

Derek Parnell wrote:
> On Fri, 22 Feb 2008 22:27:34 +0100, Don Clugston wrote:
> 
>> Janice Caron wrote:
>>> On 22/02/2008, Derek Parnell <derek at nomail.afraid.org> wrote:
>>>> Compile with the "-w" switch. It checks for 'missing' return paths.
>>> Ah great! Thanks. That will help a lot.
>>>
>>> I thought D didn't have warnings, but obviously I was wrong.
>>>
>>> It's a shame you can't /selectively/ enable warnings, because I don't
>>> actually want to be warned about "length" shadowing. Still - it's
>>> better than it's being an error. In general, I'd prefer that /all/
>>> shadowing should be allowed. But that's another debate.
>>>
>>> Anyway, thanks for the info. Will definitely be using -w from now on.
>> Really, 'missing return value' should be an error, not a warning. Unfortunately 
>> -w is horribly broken. 
> 
> The "-w" does not mean "warning" in D. Walter does not believe in warnings.

Nor do I.

> In D, it means an optional error check, that is to say, one can optionally
> choose to check for certain types of errors that are not checked for
> without the "-w" switch. I think these are optional because there is some
> debate over whether these things are actually errors or not.

The problem is that some of them have a >90% chance of being a bug, whereas 
others (such as the one I mentioned) have a ZERO % chance of being a bug.
> 
> The message text says 'warning' but the compiler treats them as errors
> anyway.
> 
>> It generates some nasty spurious errors. For example, it 
>> won't let you perform a bitwise operation on anything other than an int.
>>
>> short a;
>> short b = a | a; // won't compile with -w !!!!
> 
> Actually that turns out not to be the case. The 'warning/error' message in
> this situatation informs the coder that there is a potential loss of
> information when assigning a larger datatype to a smaller one. 
> 
> "warning - test.d(4): Error: implicit conversion of expression
>  (cast(int)a | cast (int)a) of type int to short can cause
>  loss of data"
> 
> The real problem with this is that D decides to convert your 'short' values
> to 'int' before doing the bitwise operation, leaving you with an 'int'
> value to squeeze back into a 'short'. Now you and I know that this is a
> redundant operation (converting to 'int') and that the high-end bits will
> be zeroes anyway so that there is no loss of information. But DMD is not
> that smart yet.
> 
> The 'warning' message is a valid one, in the general sense. For example
> this code here gives a similar message ...

Not really. A usually incorrect error message is much, much worse than useless.
VC6 had one which was basically "warning: you're using templates". But at least 
you could selectively turn that off. Unfortunately with this one the only option 
is to not use -w.
> 
> void main(string[] args)
> {
>   short argc = args.length;
> }
> 
> Even though we all know that we are not going to have 2^32 command line
> arguments to the program. 
> 
> The proper coding style to get the results you want is to tell the
> compiler, and other readers of your code, that you "know what you are doing
> so don't get upset"...

>  short a;
>  short b = cast(short)(a | a);
> 
>  short argc = cast(short)args.length;
> 
> These do not give 'warning' messages now.

No, but now you've introduced a latent bug! If someone changes 'a' to be (say) 
int a, it will now be accepted even though in that case it DOES involve loss of 
data! Inserting a cast is not an acceptable workaround.
Would be better in the interim to not issue 'narrowing conversion' warnings at 
all, when they are so hopelessly incorrect.