inputs skips the first file specified

[stanio]

I have a number (few hundred) of .json files which content is array (possibly empty) of objects, and I'm using the following command to merge them:

jq '[inputs | .[]]' *.json > merged.json

I've discovered the content of one of the files is missing from the result.

To Reproduce

Have three JSON files in a directory:

foo.json
bar.json
baz.json

Issue a command:

jq '[inputs | input_filename]' foo.json bar.json baz.json

The result I'm getting:

[
  "bar.json",
  "baz.json"
]

Expected behavior

A result of:

[
  "foo.json",
  "bar.json",
  "baz.json"
]

Environment (please complete the following information):

• Windows 10 64-bit (21H1 build 19043.1415)
• jq version 1.6

Additional context

I'm running in a Git-bash console:

$ jq '[inputs | input_filename]' foo.json
[]

$ jq '[inputs | input_filename]' foo.json foo.json bar.json baz.json
[
  "foo.json",
  "bar.json",
  "baz.json"
]

$ jq '[inputs | input_filename]' bar.json foo.json bar.json baz.json
[
  "foo.json",
  "bar.json",
  "baz.json"
]

input also appears to act weird:

$ jq '[input | input_filename]' foo.json
jq: error (at <unknown>): break

$ jq '[input | input_filename]' foo.json bar.json
[
  "bar.json"
]

$ jq '[input | input_filename]' foo.json bar.json baz.json
[
  "bar.json"
]
jq: error (at <unknown>): break

$ jq '[input | input_filename]' bar.json foo.json bar.json baz.json
[
  "foo.json"
]
[
  "baz.json"
]

(I'm not really sure what's the expected outcome with the last one.)

---

https://stedolan.github.io/jq/manual/v1.6/#IO

[wader]

Hey, try to add -n, otherwise jq will implicitly read one input.

$ jq ".,inputs" <(echo 1) <(echo 2)
1
2
$ jq -n "inputs" <(echo 1) <(echo 2)
1
2

Might feel be a bit unintuitive but i think it makes sense.

[stanio]

Thank you very much, @wader! Yeah, it's pretty unintuitive to me – not a jq-guru, and it gets even more confusing after reading the doc:

• --null-input/-n:

Don't read any input at all!...

😄

I confirm adding -n appears to work for me:

$ jq -n "[inputs | input_filename]" foo.json bar.json baz.json
[
  "foo.json",
  "bar.json",
  "baz.json"
]

$ jq -n '[input | input_filename]' foo.json bar.json baz.json
[
  "foo.json"
]

[wader]

Yeah the documentation could be clearer, but i understand it's a bit tricky to explain. It makes sense if you think about how jq . file works, if it did not read one input implicitly you would have to write jq "input | .." file.

How it works is that jq more or less handle all the input files as if they were one big concatenated file. Then it reads one JSON value (that can span between files) at the time until the end, and you can also explicitly read values by calling input (read one value) and inputs (read and output all remaining values one at a time). The reason i happen to know a lot about this is because i reimplement most of it for a tool based on jq :)

Here is an example explain most of it:

$ jq -n '[input,input,try input catch .]' <(echo -n 1) <(echo 2) <(echo 3)
[
  12,
  3,
  "break"
]

I recommend also look into -s (slurp) and -R (raw string input) which can be very useful.

[stanio]

Thank you again for providing an additional explanation. The issue is resolved and appears to work as designed. Please close at your discretion.

I guess an explicit reference to the -n/--null-input option could be included in the IO section of the manual as it is an option most likely (if not mandatory) needed when using input and/or inputs, as far as I understand.

[stanio]

Some alternative approaches to my original use case: Merge/concatenate arrays from multiple files

Using --slurp:

jq -s "[.[][]]" foo.json bar.json baz.json

Couldn't find a use for -R/--raw-input in this case but I hope I'll remember it when I need it. 😃

[wader]

Thank you again for providing an additional explanation. The issue is resolved and appears to work as designed. Please close at your discretion.

No problem. I think you have to close as i'm not a collaborator on this repo.

I guess an explicit reference to the -n/--null-input option could be included in the IO section of the manual as it is an option most likely (if not mandatory) needed when using input and/or inputs, as far as I understand.

Yes that would be good i think. Created #2395

[wader]

Some alternative approaches to my original use case: Merge/concatenate arrays from multiple files

Using --slurp:

jq -s "[.[][]]" foo.json bar.json baz.json

👍

If you want to be more explicit you can also use map(.[]) or maybe even just add in this case.

Couldn't find a use for -R/--raw-input in this case but I hope I'll remember it when I need it. 😃

There is also the combination -sR 😄

[stanio]

KISS:

jq -s "add" foo.json bar.json baz.json