lint: emit json

[dominikh]

We should provide a flag to output results in JSON. Such output would allow other tools to more easily consume our output.

[dmitshur]

When you say "other tools", do you mean tools that do not have a good way of parsing the file:line:col: text type of output?

[dominikh]

Well, tools like frontends shouldn't have to parse that format (which currently exists of 3 components: the location, the message, and the ID of the check). It'd be much cleaner if they could receive the information in a structured format.

[dominikh]

Preliminary design of the JSON output is a stream of JSON objects of the following kind:

{
  "checker": "staticcheck",
  "code": "SA1000",
  "severity": "warning",
  "location": {
    "file": "/some/absolute/path.go",
    "line": 123,
    "column": 123,
  },
  "message": "Your code is full of bugs"
}

Where severity is optional, and not fully defined yet, but will be used to differentiate compiler errors, suggestions (gosimple), warnings (most of staticcheck), high impact bugs and so on. We can standardize on a list later.

column is a byte offset, just like it's in the existing text output.

@bradleyfalzon as the author of gopherci, do you have any feedback on the format? any fields that are missing?

[dominikh]

Real-world example (without the severity field):

$ megacheck strings
{"checker":"staticcheck","code":"SA4017","location":{"file":"/usr/lib/go/src/strings/strings_test.go","line":954,"column":8},"message":"Repeat is a pure function but its return value is ignored"}
{"checker":"gosimple","code":"S1021","location":{"file":"/usr/lib/go/src/strings/replace_test.go","line":54,"column":2},"message":"should merge variable declaration with assignment on next line"}

[bradleyfalzon]

Looks good to me, not sure if others would prefer to handle a single json document instead of a stream.

What about errors, or is it just assumed that if you weren’t able to decode a plain text error is presented itself?

[dominikh]

not sure if others would prefer to handle a single json document instead of a stream.

It's much easier to be given a stream and treat it as a single document (read until EOF before processing) than it is to do streaming with a document.

What about errors, or is it just assumed that if you weren’t able to decode a plain text error is presented itself?

For now, if the input is malformed, we'll continue emitting errors on stderr. In the future, these may become part of the JSON.

[dominikh]

Implemented in 479b882