Allow trailing delimiters

[Nadrieril]

Fixes #512.
I discovered that the grammar allowed { , = } for an empty record literal. I'm not sure if we want to keep that.

[Gabriella439]

I would be fine with removing support for {, = } but I don't have strong opinions either way. That was an oversight from my attempt to standardize leading commas and not something I carefully thought through

[Profpatsch]

Doesn’t allowing a trailing delimiter require backtracking all the way in parsers? I implemented that for the nix parser some time ago, but we scrapped it because of that.

[Gabriella439]

@Profpatsch: I believe it's still doable without full backtracking. Using commas as an example, after parsing any given comma-separated element you parse one of:

    try ("," *> "})
<|> "}"
<|> ("," *> parseElement *> parseRest)

... which limits backtracking to a limited number of characters

[Nadrieril]

Isn't that lookahead rather than backtracking then ? I don't remember that being a performance sink

[Nadrieril]

Does anyone object to this then ?

[philandstuff]

Can we have tests for the case of having both a leading and a trailing delimiter? eg [ , 3 , ] is valid according to the new ABNF.

[Gabriella439]

@Nadrieril: I still think this proposal is a good idea

[Gabriella439]

I think this is ready to be merged

[Nadrieril]

Yeah sorry I haven't had the bandwidth for dhall recently, and I wanted to add the tests that @philandstuff suggested