Remove matching/non-matching elements of a nested array using jq

Question

I need to split the results of a sonarqube analysis history into individual files. Assuming a starting input below,

    {
  "paging": {
    "pageIndex": 1,
    "pageSize": 100,
    "total": 3
  },
  "measures": [
    {
      "metric": "coverage",
      "history": [
        {
          "date": "2018-11-18T12:37:08+0000",
          "value": "100.0"
        },
        {
          "date": "2018-11-21T12:22:39+0000",
          "value": "100.0"
        },
        {
          "date": "2018-11-21T13:09:02+0000",
          "value": "100.0"
        }
      ]
    },
    {
      "metric": "bugs",
      "history": [
        {
          "date": "2018-11-18T12:37:08+0000",
          "value": "0"
        },
        {
          "date": "2018-11-21T12:22:39+0000",
          "value": "0"
        },
        {
          "date": "2018-11-21T13:09:02+0000",
          "value": "0"
        }
      ]
    },
    {
      "metric": "vulnerabilities",
      "history": [
        {
          "date": "2018-11-18T12:37:08+0000",
          "value": "0"
        },
        {
          "date": "2018-11-21T12:22:39+0000",
          "value": "0"
        },
        {
          "date": "2018-11-21T13:09:02+0000",
          "value": "0"
        }
      ]
    }
  ]
}

How do I use jq to clean the results so it only retains the history array entries for each element? The desired output is something like this (output-20181118123808.json for analysis done on "2018-11-18T12:37:08+0000"):

{
  "paging": {
    "pageIndex": 1,
    "pageSize": 100,
    "total": 3
  },
  "measures": [
    {
      "metric": "coverage",
      "history": [
        {
          "date": "2018-11-18T12:37:08+0000",
          "value": "100.0"
        }
      ]
    },
    {
      "metric": "bugs",
      "history": [
        {
          "date": "2018-11-18T12:37:08+0000",
          "value": "0"
        }
      ]
    },
    {
      "metric": "vulnerabilities",
      "history": [
        {
          "date": "2018-11-18T12:37:08+0000",
          "value": "0"
        }
      ]
    }
  ]
}

I am lost on how to operate only on the sub-elements while leaving the parent structure intact. The naming of the JSON file is going to be handled externally from the jq utility. The sample data provided will be split into 3 files. Some other input can have a variable number of entries, some may be up to 10000. Thanks.

Answer 1

Here is a solution which uses awk to write the distinct files. The solution assumes that the dates for each measure are the same and in the same order, but imposes no limit on the number of distinct dates, or the number of distinct measures.

jq -c 'range(0; .measures[0].history|length) as $i
  | (.measures[0].history[$i].date|gsub("[^0-9]";"")),  # basis of filename
    reduce range(0; .measures|length) as $j (.;
      .measures[$j].history |= [.[$i]])' input.json |
awk -F\\t 'fn {print >> fn; fn="";next}{fn="output-" $1 ".json"}'

Comments

The choice of awk here is just for convenience.

The disadvantage of this approach is that if each file is to be neatly formatted, an additional run of a pretty-printer (such as jq) would be required for each file. Thus, if the output in each file is required to be neat, a case could be made for running jq once for each date, thus obviating the need for the post-processing ( awk ) step.

If the dates of the measures are not in lock-step, then the same approach as above could still be used, but of course the gathering of the dates and the corresponding measures would have to be done differently.

Output

The first two lines produced by the invocation of jq above are as follows:

"201811181237080000"
{"paging":{"pageIndex":1,"pageSize":100,"total":3},"measures":[{"metric":"coverage","history":[{"date":"2018-11-18T12:37:08+0000","value":"100.0"}]},{"metric":"bugs","history":[{"date":"2018-11-18T12:37:08+0000","value":"0"}]},{"metric":"vulnerabilities","history":[{"date":"2018-11-18T12:37:08+0000","value":"0"}]}]}

Answer 2

In the comments, the following addendum to the original question appeared:

is there a variation wherein the filtering is based on the date value and not the position? It is not guaranteed that the order will be the same or the number of elements in each metric is going to be the same (ie some dates may be missing "bugs", some might have additional metric such as "complexity").

The following will produce a stream of JSON objects, one per date. This stream can be annotated with the date as per my previous answer, which shows how to use these annotations to create the various files. For ease of understanding, we use two helper functions:

def dates:
  INDEX(.measures[].history[].date; .)
  | keys;

def gather($date): map(select(.date==$date));

dates[] as $date
| .measures |= map( .history |= gather($date) )

INDEX/2

If your jq does not have INDEX/2 , now would be an excellent time to upgrade, but in case that's not feasible, here is its def:

def INDEX(stream; idx_expr):
  reduce stream as $row ({};
    .[$row|idx_expr|
      if type != "string" then tojson
      else .
      end] |= $row);