Possibility to render quoted keys?

[markus1189]

In dhall-lang/dhall-json#41 there is a problem with quoting keys of objects using the yaml library.

The issue is that string keys are never quoted which might lead to unintentional conversion. For example if you convert a string key "NO" to a yaml like this:

NO: Norway

the NO is now a boolean in YAML and no longer a string. Is there anyway to force the quoting of the key in that case?

[snoyberg]

I'm not opposed to this in theory, we already use such logic for string values of mappings. But I'd like to know where exactly this causes breakage. AIUI, the quoting of strings is never strictly necessary according to the YAML spec, it's only used because some interpretations (including ours!) will eagerly convert from string into a different scalar for values to make things a bit more type-safe and/or match JSON semantics. However, I'm unaware of any library doing the same thing with keys.

[markus1189]

My use case is that I have a Dhall record like this:

{ NO: "Norway"
, DE: "Germany"
}

when converting it to yaml using this library, the problem is that NO will be a boolean in the resulting YAML, while DE is still (correctly) a string:

NO: Norway
DE: Germany

To fix this accidental conversion it would be great if one could specify that keys should be quoted.

[snoyberg]

Which library is interpreting the YAML that way? It's definitely not the yaml library itself, since it's mapping to JSON it will always treat keys as string scalars:

#!/usr/bin/env stack
-- stack --resolver lts-11.10 script
{-# LANGUAGE OverloadedStrings #-}
import Control.Exception
import Data.Yaml

main :: IO ()
main = do
  val <- either throwIO pure
       $ decodeEither' "NO: Norway\nDE: Germany\n"
  print (val :: Value)

Results in:

$ stack Main.hs
Object (fromList [("DE",String "Germany"),("NO",String "Norway")])

[markus1189]

According to the YAML spec, NO is a boolean, any implementation that adheres to that should interpret

NO: Norway
DE: Boolean

as

false: "Norway"
"DE": "Germany"

Examples that do this: http://www.yamllint.com/, https://yaml-online-parser.appspot.com/ and other implementations like the standard ruby yaml library:

irb(main):001:0> require 'yaml'                                                                                                                                                                                    
=> true                                                                                                                                                                                                            
irb(main):002:0> YAML.load_file('/tmp/norway.yaml')                                                                                                                                                                
=> {false=>"Norway"}

and also pythons pyyaml:

>>> import yaml                                                                                                                                                                                                    
>>> yaml.load(open('/tmp/norway.yaml'))                                                                                                                                                                            
{False: 'Norway'}

(Btw, yes, YAML is weird... 😂 )

[snoyberg]

Alright, I can confirm you're right. Initially, I had been under the impression that the parser had the prerogative to "cast" scalars to either text, bool, or numeric, depending on what succeeded. It seems I was quite mistaken about that, which in turn means that our handling of scalar values is more correct than I realized. On the other hand, quoting the scalar keys is necessary, as you've pointed out. I'll push a commit in a bit.

[hvr]

@markus1189 btw, re "According to the YAML spec, NO is a boolean" -- that's not a true statement; NO definitely not a valid boolean (i.e. tag:yaml.org,2002:bool) representation in the most recent http://yaml.org/spec/1.2/spec.html

I recommend https://hackage.haskell.org/package/HsYAML if you need a standards-compliant YAML parser.

[markus1189]

You're right, seems like that changed. I was referring to http://yaml.org/type/bool.html

…

On Mon, Jul 9, 2018, 15:39 Herbert Valerio Riedel ***@***.***> wrote: @markus1189 <https://github.com/markus1189> btw, re "According to the YAML spec, NO is a boolean" -- that's not a true statement; NO definitely *not* a boolean in the most recent http://yaml.org/spec/1.2/spec.html — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#137 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAkGz64-71TiRPzgC1NdymY6jw8y0TJjks5uE10KgaJpZM4VGo3u> .

[hvr]

@markus1189 yeah, that spec you're referring to was merely a working draft

[markus1189]

Strangely enough it seems like most implementations allow those forms for booleans, even though it was only a draft. Really confusing

…

On Mon, Jul 9, 2018, 15:43 Herbert Valerio Riedel ***@***.***> wrote: @markus1189 <https://github.com/markus1189> yeah, that spec you're referring to was merely a working draft — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#137 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAkGz38yO5UBql5KutQOtV9ilgXU3Uejks5uE133gaJpZM4VGo3u> .

[hvr]

@markus1189 It's a long story... basically those aren't YAML 1.2 (released in 2009) compliant implementations but may only be implementing the ancient YAML 1.1 (released in 2005) spec. I'm not sure what yaml implements exactly, but it's certainly isn't YAML 1.2 (HsYAML however implements YAML 1.2).

[snoyberg]

Enough @hvr, no more spam.