compile a.0.1 to a[0][1]

[michaelficarra]

Idea from coco, implemented at satyr/coco@c395b035

Anyone against this? It seems like there's no downside, it's convenient, and it looks nicer.

[akidee]

Good idea

[Sh4rK]

I agree

[odf]

I think it looks confusing.
coffee> a = { '0.1': 'point one' }
{ '0.1': 'point one' }
coffee> a[0.1]
point one

[TrevorBurnham]

Yes, +1 from me as well. odf raises a good point about a.0.1 feeling ambiguous if a[0.1] is defined, but keys with .s in them are pretty uncommon. In the much more common case—arr.0, arr.1, etc.—the dot syntax helps to reduce punctuation noise.

[satyr]

Note that Coco also allows:

a.'0'
a."#{1-1}"

[odf]

As I remarked elsewhere, I would also like to be able to write things like a.+ instead of a["+"]. That would save me three characters each time instead of just one, and get me closer to the conventional notation instead of further away from it. That doesn't necessarily mean I'm against this suggestion, but maybe it would be useful to think a bit more broadly. What can we reasonably put after a dot and expect coffeescript to interpret it as an attribute name?

[TrevorBurnham]

I'm not sure about a.+, and even a.0.1 makes me a little hesitant. But strings, as in satyr's examples above, seem like a slam dunk to me. Square brackets should be reserved for cases where the key is variable. (Note that a[0] and a['0'] are equivalent; keys are always coerced to strings.) So I think that allowing a.'str'/a."str" would be a pretty conservative, readability-boosting addition to the language.

[michaelficarra]

@TrevorBurnham: While I don't disagree, I'd say that is probably better off as a separate proposal. Let's keep this discussion to the originally-proposed a.2.15 syntax where integers alone are allowed to be used as property names.

[odf]

If it's all about reducing punctuation, wouldn't a[2,15] look more natural and less confusing than a.2.15?

[TrevorBurnham]

That's actually a really interesting idea, though I'm hesitant to yet another meaning to the once-innocuous string of characters [a, b]...

[satyr]

Note also that the original commit that implemented this is satyr/coco@b368088c. The one in OP is a bug fix for it.

[jashkenas]

I'm afraid I'm rather strongly against this, as it divorces what can be used as dotted accessors from what can be used as a valid variable name. Right now, for any x that you can write obj.x = val, you can also write x = val. After this change, that's no longer the case. Closing as a wontfix.

[satyr]

for any x

no

[michaelficarra]

satyr's got you on that one

[akidee]

obj['x'] = val

and

['x'] = val

Alternative:

'x' = val

But how can this be translated into JS? I think the only way is eval()