On Sunday, 13 September 2020 at 09:40:34 UTC, claptrap wrote:
> Maybe the problem isnot enough exploratory work is done up front?

Various forms of this have come up over the years.

A PR in 2017 offered just the basics done automatically: https://github.com/dlang/dmd/pull/6703

This kind of thing is perhaps the most obvious - just have the compiler rewrite it for you. Very limited, though, it only really works for the one use case. Been discussed several times before this too, but never went anywhere.

This one in 2018 though opened up some new ideas, from Jonathan Marler:
https://github.com/dlang/dmd/pull/7988

The key innovation there was instead of calling a function, it just built an argument list. This is when these proposals actually started getting interesting to me: it makes it usable in a lot more cases, though the cost is it isn't as convenient. It is very simple though.

As you can see though, it got a great many comments and even a DIP draft <https://github.com/dlang/DIPs/pull/140/files>. In 2019, Walter started thinking in terms of printf, and this eventually led to his DIP 1027, which had the compiler build a string and reorder the arguments for you. It is still a parameter list you just pass to another function, but it is partially cooked. This adds to the complexity but opened up more possibilities.

That's an interesting innovation too, but his DIP had a number of flaws and missed potential which this new DIP aims to patch. The major innovation here is adding a new type to the tuple.

Over the last couple days, we've had some new ideas came up and I considered what we could do with them.

Of course, there's a lot of other languages with things too, C# and Javascript both have quite interesting offerings, if you haven't looked at them, you should. The links are in the OP's references.


Anyway, meanwhile, Steven went back to the old ideas. So instead of adding new stuff, he shattered it and picked up just the pieces that survived. And what we're looking at now just takes the best one feature from each generation and puts it together. Convenience of gen one, flexibility of gen two, power of gen three. Without the rest of the accumulated cruft.

At first, I pushed back at him saying various use cases wouldn't work... but turns out they do. Much of the other stuff genuinely wasn't necessary. So we wrote the first several drafts and now are in the editing process to pare it back down.

> IE. Maybe have 4 or 5 people come up with as many solutions as possible, literally just write down ideas as they come to you. And then bring them all together and evaluate them. It's usually the outlier ideas that can spur people to "think differently" about the problem even if those outliers dont end up going anywhere.

Indeed, I often "think out loud" in these thread. Most of it goes nowhere... but every so often some piece of it endures. If we just instantly shot everything down instantly, we'd flush the gold with the dirt.

On Monday, 14 September 2020 at 02:54:35 UTC, Adam D. Ruppe wrote:
> On Sunday, 13 September 2020 at 09:40:34 UTC, claptrap wrote:
>>
>> IE. Maybe have 4 or 5 people come up with as many solutions as possible, literally just write down ideas as they come to you. And then bring them all together and evaluate them. It's usually the outlier ideas that can spur people to "think differently" about the problem even if those outliers dont end up going anywhere.
>
> Indeed, I often "think out loud" in these thread. Most of it goes nowhere... but every so often some piece of it endures. If we just instantly shot everything down instantly, we'd flush the gold with the dirt.

That's still not the same thing. These threads are full of assessment and evaluation. The first reply to one of your ideas will be someone evaluating it. That's not good for spurring more ideas. Think like improv comedy, if after each time someone said something there was a pause for evaluation of how funny it was, it would kill the momentum.

To really broaden the idea search space there needs to be no evaluation while you're generating ideas.

If that was done up front, you might have skipped 2 years of DIPs that went nowhere? I dont know maybe there are constraints on language design that means theres not a large space to explore.

On Tuesday, 15 September 2020 at 09:30:37 UTC, claptrap wrote:
> These threads are full of assessment and evaluation.

Yeah, it kinda sucks.

> If that was done up front, you might have skipped 2 years of DIPs that went nowhere?

There's only actually been two formal DIPs: Walter's and this one, which is just a tweak to Walter's (though the next version after this round is a full rewrite).

The rest of it has been sample implementations or just informal talk. Just it is worth noting this conversion has been going on for a while including a lot more people than are present in this thread.