Re: Expression-local variables and 'let'

Isaac Gouy wrote:

>>After all, what matters is the abstract syntax. Tools should be able
>>to understand and present you with the concrete syntax you personally
>>feel more comfortable with.
>>    
>>
>
>Isn't language design more interesting than that?
>Less about preferences and comfort levels, more about providing
>structures that buffer against slips and mistakes.
>  
>
I'm not saying syntax design is not interesting. I'm saying that there 
is more to language design than syntax design (there are the semantic 
constructs you provide, type checking, ...), and that the syntax part 
can be treated (mostly) independently from the rest.

>Isn't part of what Nice is about safety?
>  
>
Yes.

>IIRC one of the unsafe aspects of imperative languages is that it isn't
>clear when we've changed the state - so we've had a whole slew of
>single assignment languages.
>
>What was the objection to making single-assignment default?
>  
>
A question of priorities. Nice has a few strong points to make. One is 
that a powerful type system can be used in practice, and that it bring 
improtant benefits. There are other interesting type systems around, but 
Nice's is unique.

A second is that multi-methods is a better approach than class-based OO. 
There are other languages with multi-methods, but no mainstream one, so 
this point is also an important one to make.

Third, that this approach to "OO" fits particularly well with the 
functional paradigm.

2 and 3 taken together mean that you can basically get the best of both 
the OO and the FP world.

There are other points, like a better object creation model, avoiding 
boiler-plate code like getters and setters (in preparation), contracts 
(just applying previous work here), and a few more.

All these taken together are already quite a lot to digest for 
"mainstream".

My goal with Nice is to demonstrate in practice that the above main 
points are worth learning about, and to provide a very pleasant language 
to work with.

Now, we could discuss forever about the ideal form of every single 
detail of a language. The problem is that there are many cases where an 
objective ideal is not something people will agree on, or where that 
ideal might be only a minor improvement over a much more known practice, 
in which case experience shows that it will probably not catch on.

In those cases, I think it's the best choice to stick with something 
people are familiar with, or will easily relate to, unless there is a 
compelling reason not to do so (that's how new main points can be added 
to the list). This approach ensures that we don't hide our main points 
behind details which will cause pain to many users, and repell many others.

Changing the default mutability status, for instance, seems to me to 
fall into this category. It might be a slight improvement, but the 
benefits would not be huge, especially as you kep both option (single 
assignment languages or pure functional languages are the ones that 
really explore a different way here). Anyway, with let and var, there is 
no default anymore, so you can just express your intention.

Going back to generalities, I understand that this pragmatic approach 
can frustrate some people (those who speak up ;-) who already got the 
main points, who have a large culture about programming languages, and 
who would like more radicality. The thing is, we also need to keep the 
general picture in mind. I'd prefer a language that brings novel and 
little-known ideas to a larger community (which itself brings the 
perspective of more libraries, tools, ...) than a "perfect" language 
that will appeal to very few because it's too big of a step for most 
people to make. This might be were an alternate syntax comes in, if 
really this is an important issue to some.

I'm not opposed to breaking conventions. I just want to make sure the 
benefit is clearly worth the extra "peculiarity".

Daniel



-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click