Empty environment variable problem

[Pitometsu]

How to pass an environment variable which is set, but value is empty, and expected type is string (Text)? (shoul be encoded as ''value in yaml configuration)
Currently it interprets empty environment variable as a null (just empty space in yaml configuration, of Null type).

app: Could not convert to expected type: Error in $.field.name: parsing Text failed, expected String, but encountered Null
CallStack (from HasCallStack):
  error, called at src/Data/Yaml/Config.hs:199:19 in yaml-0.11.2.0-44sOdEiCtMxGtgOoTNcdTZ:Data.Yaml.Config
app: (exit 1): failed

[Pitometsu]

workaround

---

fail:

field:
  name: _env:ENV

but this works:

field:
  name: _env:ENV:''

[snoyberg]

The comment in the code you link seems to describe the situation. It's interpreting whether to treat as a string or not based on the default value. And otherwise it parses the input as YAML, which for an empty string is in fact null I believe.

[Pitometsu]

Can similar behavior for Text be added? Not sure about ByteString. Thank you for pointing to the comment.

[snoyberg]

I don't understand the request. This is working as designed. Please clarify what alternative implementation you'd like to see, "similar behavior" doesn't make sense in this context, at least to me.

[Pitometsu]

I mean: e.g. read empty environment variable value for all IsString instances (or any custom inherited class instances) as a '' not null, not String only, to make it slightly more open for extension. Would that make sense?

[snoyberg]

That would break existing programs, which expect YAML parsing to occur.

[Pitometsu]

Reasonable remark, that's sad. But there still can be added class IsString t => AsString t.
And doing so for all the AsString. And derived instance for Text (e.g. in distinct module that supposed to be imported explicitly to avoid breaking the exist behavior).

[Pitometsu]

Wait a moment. Which cases will that brake? Currently the lib just generate a runtime error for null value, which is not instance of Text. It just expected that library will successfully parse the valid value of empty environment variable to appropriated value of Text type, as it intended to. I see no breaking changes required.

[snoyberg]

This is all too vague for me to follow. I can’t guarantee it will get merged, but at this point I think your best bet is opening a PR with the changes you want to see.

[sjshuck]

set, but value is empty

read empty environment variable value

successfully parse the valid value of empty environment variable

This makes no sense on Windows because there is no difference between an unset environment variable and one set to the empty string, even at the C API level.