Allow parsing a wider range of keys?

[jaklt]

In the newest version of stache a wider range of keys is allowed to be parsed in order to better comply with the mustache documentation. In particular, one of the examples in the documentation reads as

{{#person?}}
  Hi {{name}}!
{{/person?}}

Both haskell packages stache and mustache (as well as the official documentation) recognise that the outer block has a valid key person?. See also issues mrkkrp/stache#53 and mrkkrp/stache#54 for details.

[phadej]

There isn't falling cases in specification. Is there cases for such keys in recent versions of https://github.com/mustache/spec/? Does spec has non-example definition of key grammar?

Otherwise i'm tempted to simply add ? to recognized characters, that would be example complying and simple!

[jaklt]

The issue is that the specification is silent about this. Or at least I couldn't find any mention of this.

Look for example what https://mustache.github.io/#demo allows to parse. Given the mustache source

{{# c!@#$%^&( . )[] <> / }}
{{   a.b   }}
{{  a .b    }}
{{  a . b    }}
{{/ c!@#$%^&( . )[] <> /   }}

and the JSON data

{
  "c!@#$%^&( ": {
    " )[] <> /": true
  },
  "a": { "b": "foo" },
  "a ": { "b": "bar", " b": "baz"  }
}

it produces

foo
bar
baz

The advantage of allowing a wider range of characters in the keys (including whitespaces in the middle) is that then expressions like {{ max(a,b) }} can be parsed. Then such expression can be reduced by pefrorming an operation on the AST (and evaluated as the max of the two values). Note that this is not possible with lambdas because those are only unary functions.

[phadej]

max(a,b) is against the spirit of logicless templating mustache is.

I don't care what other implementation do, if doesn't require some input to be accepted (or rejected), i as an implementor have a choice, and I make conservative "javascript identifier" choice (not really, I just copied what stache did at the time of the fork).

In short, adding ? can be argued for, as that example is in the official examples, however it's not in the official specs, so it can be argued that example is wrong and https://mustache.github.io/#demo implementation is lax.

That is something formustache/spec to clarify.

---

In general, my approach to imprecisely specified formats is to do as little as possible. Accepting more feels wrong. It is my opinion.

[jaklt]

That makes sense, I understand the sentiment of implementing as little as possible whenever the specification is too vague. I might try to clarify this with the spec team.

I think that with the max(a,b) example, you're not allowing "more logic" in the templates with this than you're already allowed to do with Lambdas. The slogan "logic-less templates" is great but (luckily) not fully fulfilled in mustache. It's probably not worth debating this here, it depends on everybody's point of view.