I was just thinking about porting a javascript function to
D that takes an object of params:

function foo(args) { ... }

foo( { "closeButton" : false, "width" : 200 } );


In D, I figured I'd make args a struct:

struct FooArgs {
   bool closeButton = true; // the default is for it to be there
   int width = 600;
}

void foo(FooArgs args) { ... }


And, of course, this works, but it was leaving a declared
variable around and can be somewhat repetitive. But first
I tried making it look kinda like the JS:

FooArgs args = { closeButton: false };
foo(args);

But, that style struct initializer doesn't respect the default
values. Here, width == 0. No good.


So, then, well, that's OK, I'll just write it out:

FooArgs args;
args.closeButton = false;
foo(args);



There we go, not too bad at all. We can kill some repetition
of "args" with the with() {} statement. Great!

But, what if I have another call later? I don't want that args
sticking around.


FooArgs args;
// ...
foo(args);

BarArgs args; // error, args already declared




No prob, let's scope it! I tried:


with(FooArgs args) {
   closeButton = false;
   foo(args);
}

// can redeclare another args here if you want


But it didn't work. It is with(symbol) or with(expression)
in the spec.



Would it work to also allow with(decl)? It seems to me that
it should be ok... kinda similar to the currently allowed

if(auto a = foo()) { /* use a here */ }


So we're just extending that same idea out to the with statement
too.

On 07/23/12 16:37, Adam D. Ruppe wrote:
> But it didn't work. It is with(symbol) or with(expression)
> in the spec.
> 
> 
> 
> Would it work to also allow with(decl)? It seems to me that it should be ok... kinda similar to the currently allowed
> 
> if(auto a = foo()) { /* use a here */ }
> 
> 
> So we're just extending that same idea out to the with statement too.

The more general form would be to make variable declaration an expression. I'm not sure if that causes any ambiguity, parsing-wise, but it allows other neat things as well, like while(vardecl) and if (decl1 && condition).

On Monday, 23 July 2012 at 14:46:30 UTC, FeepingCreature wrote:
> The more general form would be to make variable declaration an expression.

Right, and that would be pretty amazing, but it would probably
be too hard to do in D today...

Le 23/07/2012 16:37, Adam D. Ruppe a écrit :
> I was just thinking about porting a javascript function to
> D that takes an object of params:
>

Let's not add another special case for the declaration can be used here instead of an expression. This is already the case in if and it will proliferate.

The real need here is to allow variable declaration to be used as expressions. And functions too as they are first class citizens now.

This reduces special cases instead of adding new one, and give a much greater opportunity for expressiveness in the language.

Le 23/07/2012 16:46, FeepingCreature a écrit :
> On 07/23/12 16:37, Adam D. Ruppe wrote:
>> But it didn't work. It is with(symbol) or with(expression)
>> in the spec.
>>
>>
>>
>> Would it work to also allow with(decl)? It seems to me that
>> it should be ok... kinda similar to the currently allowed
>>
>> if(auto a = foo()) { /* use a here */ }
>>
>>
>> So we're just extending that same idea out to the with statement
>> too.
>
> The more general form would be to make variable declaration an expression. I'm not sure if that causes any ambiguity, parsing-wise, but it allows other neat things as well, like while(vardecl) and if (decl1&&  condition).

With the current grammar, it is difficult, because declarations are statements.

On Monday, 23 July 2012 at 15:37:22 UTC, deadalnix wrote:
> The real need here is to allow variable declaration to be used as expressions. And functions too as they are first class citizens now.

I don't think that's actually going to happen though.
It is just a big, big change.

Adding with(decl) is a small change. I think anyway.

> This reduces special cases instead of adding new one, and give a much greater opportunity for expressiveness in the language.

I agree this would be nice, but I'm just trying to ask
for something I think we can actually get...

I can't help thinking it sounds rather like a job for... named parameters.  Just imagine it:

##############################
void foo ( bool closeButton = true, int width = 600 ) {
    writeln( closeButton, ", ", width );
}

void main () {
    foo( closeButton: false );
}
##############################

With output:
false, 600

Gosh, that'd sure be swell.  (Either that or find a way to sanely allow non-static struct initializers, at least for rvalue arguments.)

-- Chris NS

On 23/07/12 17:04, Adam D. Ruppe wrote:
> On Monday, 23 July 2012 at 14:46:30 UTC, FeepingCreature wrote:
>> The more general form would be to make variable declaration an
>> expression.
>
> Right, and that would be pretty amazing, but it would probably
> be too hard to do in D today...


The bizarre thing, however, is that the compiler uses it internally. There's thing called a DeclarationExp. It's created all the time when lowering occurs.
It"s only in the parser that declarations are not valid as expressions.

On Tuesday, 24 July 2012 at 07:03:05 UTC, Chris NS wrote:
> I can't help thinking it sounds rather like a job for... named parameters.  Just imagine it:

Yeah, that could do it too, but named parameters have been
brought up a few times too and there's opposition to it.

I kinda prefer the struct to named params anyway because
you can store it for later too. But I could go either way.

On Tuesday, 24 July 2012 at 23:03:39 UTC, Adam D. Ruppe wrote:
> On Tuesday, 24 July 2012 at 07:03:05 UTC, Chris NS wrote:
>> I can't help thinking it sounds rather like a job for... named parameters.  Just imagine it:
>
> Yeah, that could do it too, but named parameters have been
> brought up a few times too and there's opposition to it.
>
> I kinda prefer the struct to named params anyway because
> you can store it for later too. But I could go either way.

Oh I know; which is why I wrote in a slightly snarky manner.  I still hold out hope.  Your reuse argument for structs is, of course, completely valid and compelling.

Another possibility, in cases like that presented in the original post, is to write a single struct to be used with all the relevant functions -- but then there's the problem of memory abuse.  In some cases it may be fine, but tossing around several copies of a very large struct, and only using two or three of the fields in a given case, is just unreasonable.

-- Chris NS