Investigate spurious `[]` on JSON output in some cases about miller HOT 4 CLOSED

Just thinking out loud here… Miller is pretty forgiving about JSON input. All of these are interpreted as two records:

# One JSON document, an array of two objects
echo '[{"foo":"bar"},{"foo":"baz"}]' | mlr -j cat
# "JSON Lines" - two JSON documents, one object each
echo -e '{"foo":"bar"}\n{"foo":"baz"}' | mlr -j cat
# Neither, but still accepted
echo '{"foo":"bar"}{"foo":"baz"}' | mlr -j cat

The output is normally a single JSON document consisting of an array, but if --no-jlistwrap is set, it's a JSON Lines document.

I wonder if the JSON input reader could set --no-jlistwrap if it encountered input other than a JSON array?

from miller.

So @smammy ... it turns out this was intentional on my part.

Here:
https://github.com/johnkerl/miller/blob/6.11.0/pkg/output/record_writer_json.go#L56-L88
the idea is that mlr --json cat /dev/null should produce [].

However -- I am questioning that. For example, jq . /dev/null produces nothing at all and ostensibly Miller shouldn't either. 🤔

from miller.

Also:

$ echo '' | jq .

$ echo '[]' | jq .
[]

$ echo '' | mlr --json cat
[
]

$ echo '[]' | mlr --json cat
[
]

It's easy for me to make those last two both empty, or both [] ... issue is, Miller "forgets" the outer [] provided on input, even when it's explicitly provided as input, so:

• I don't like empty input leading to [] output
• I don't like [] input leading to empty output
• I'm not seeing a clean way for the JSON-output writer to know which came in in the first place ...

from miller.

@smammy yes, I think something like that.

I don't want to mutate CLI arguments per se, but, in the context struct that comes in with each record, I can put a JSON-had-brackets flag. So the JSON record-reader can set this on the end-of-stream marker, and the JSON record-writer can consult it, and all other format-specific record-readers and record-writers can ignore it.

from miller.