Shell Script – How to Parse Stdout as a Mix of CSV and JSON

awkcsvjsonshell-scripttext processing

I'm currently working on a class that has us submit our code to an auto-grader and then returns our results. The format it returns is kind of hard to visually parse, so I would like to write a script that I can use in a pipe to make it easier to read.

Here is the output of the auto-grader:

Problem,Correct?,Correct Answer,Agent's Answer
"Challenge Problem B-04",0,4,-1
"Basic Problem B-12",0,1,-1
"Challenge Problem B-05",0,6,-1
"Challenge Problem B-07",0,6,-1
"Challenge Problem B-06",0,3,-1
"Basic Problem B-11",0,1,-1
"Basic Problem B-10",0,3,-1
"Challenge Problem B-03",0,3,-1
"Challenge Problem B-02",0,1,-1
"Challenge Problem B-01",0,6,-1
"Challenge Problem B-09",0,4,-1
"Challenge Problem B-08",0,4,-1
"Basic Problem B-08",0,6,-1
"Basic Problem B-09",0,5,-1
"Basic Problem B-04",0,3,-1
"Basic Problem B-05",0,4,-1
"Basic Problem B-06",0,5,-1
"Basic Problem B-07",0,6,-1
"Basic Problem B-01",0,2,-1
"Basic Problem B-02",0,5,-1
"Basic Problem B-03",0,1,-1
"Challenge Problem B-10",0,4,-1
"Challenge Problem B-11",0,5,-1
"Challenge Problem B-12",0,1,-1
{
    "Basic Problems B": {
        "Incorrect": "0",
        "Skipped": "12",
        "Correct": "0",
        "Set": "Basic Problems B"
    },
    "Challenge Problems B": {
        "Incorrect": "0",
        "Skipped": "12",
        "Correct": "0",
        "Set": "Challenge Problems B"
    }
}

It's a mix of comma separated values and JSON. It would be nice to place this all in a pretty table that I can read.

Currently, I have something like

python submit.py --provider gt --assignment error-check | column -t -s, | less -S

Which outputs:

{
    "Basic Problems B": {
        "Incorrect": "0",
        "Skipped": "12",
        "Correct": "0",
        "Set": "Basic Problems B"
    },
    "Challenge Problems B": {
        "Incorrect": "0",
        "Skipped": "12",
        "Correct": "0",
        "Set": "Challenge Problems B"
    }
}
Problem                   Correct?  Correct Answer  Agent's Answer
"Challenge Problem B-04"  0         4               -1
"Basic Problem B-12"      0         1               -1
"Challenge Problem B-05"  0         6               -1
"Challenge Problem B-07"  0         6               -1
"Challenge Problem B-06"  0         3               -1
"Basic Problem B-11"      0         1               -1
"Basic Problem B-10"      0         3               -1
"Challenge Problem B-03"  0         3               -1
"Challenge Problem B-02"  0         1               -1
"Challenge Problem B-01"  0         6               -1
"Challenge Problem B-09"  0         4               -1
"Challenge Problem B-08"  0         4               -1
"Basic Problem B-08"      0         6               -1
"Basic Problem B-09"      0         5               -1
"Basic Problem B-04"      0         3               -1
"Basic Problem B-05"      0         4               -1
"Basic Problem B-06"      0         5               -1
"Basic Problem B-07"      0         6               -1
"Basic Problem B-01"      0         2               -1
"Basic Problem B-02"      0         5               -1
"Basic Problem B-03"      0         1               -1
"Challenge Problem B-10"  0         4               -1
"Challenge Problem B-11"  0         5               -1
"Challenge Problem B-12"  0         1               -1

This gets me most of the way there. Now I'm wondering if there is a way I can handle the JSON?

I can't rely on splitting the output at a certain line-number, but am thinking I can segment the output when it first finds a {.

I would like to do this as minimal as possible so I can share it with classmates. So the less dependencies the better.

I've seen other JSON parsing posts suggesting to use external codes.

An ideal output would look something like:

Problem                   Correct?  Correct Answer  Agent's Answer
"Challenge Problem B-04"  0         4               -1
"Basic Problem B-12"      0         1               -1
"Challenge Problem B-05"  0         6               -1
"Challenge Problem B-07"  0         6               -1
"Challenge Problem B-06"  0         3               -1
"Basic Problem B-11"      0         1               -1
"Basic Problem B-10"      0         3               -1
"Challenge Problem B-03"  0         3               -1
"Challenge Problem B-02"  0         1               -1
"Challenge Problem B-01"  0         6               -1
"Challenge Problem B-09"  0         4               -1
"Challenge Problem B-08"  0         4               -1
"Basic Problem B-08"      0         6               -1
"Basic Problem B-09"      0         5               -1
"Basic Problem B-04"      0         3               -1
"Basic Problem B-05"      0         4               -1
"Basic Problem B-06"      0         5               -1
"Basic Problem B-07"      0         6               -1
"Basic Problem B-01"      0         2               -1
"Basic Problem B-02"      0         5               -1
"Basic Problem B-03"      0         1               -1
"Challenge Problem B-10"  0         4               -1
"Challenge Problem B-11"  0         5               -1
"Challenge Problem B-12"  0         1               -1

Set                   Incorrect Skipped Correct
Basic Problems B      0         12      0
Challenge Problems B  0         12      0

Best Answer

Separating the JSON from the rest is quite easy. This will give you the non JSON only:

python submit.py --provider gt --assignment error-check | sed '/{/,$d'

And this, only the JSON:

python submit.py --provider gt --assignment error-check | sed -n '/{/,$p'

To illustrate, I have saved your example input as file and:

$ sed '/{/,$d' file
Problem,Correct?,Correct Answer,Agent's Answer
"Challenge Problem B-04",0,4,-1
"Basic Problem B-12",0,1,-1
"Challenge Problem B-05",0,6,-1
"Challenge Problem B-07",0,6,-1
"Challenge Problem B-06",0,3,-1
"Basic Problem B-11",0,1,-1
"Basic Problem B-10",0,3,-1
"Challenge Problem B-03",0,3,-1
"Challenge Problem B-02",0,1,-1
"Challenge Problem B-01",0,6,-1
"Challenge Problem B-09",0,4,-1
"Challenge Problem B-08",0,4,-1
"Basic Problem B-08",0,6,-1
"Basic Problem B-09",0,5,-1
"Basic Problem B-04",0,3,-1
"Basic Problem B-05",0,4,-1
"Basic Problem B-06",0,5,-1
"Basic Problem B-07",0,6,-1
"Basic Problem B-01",0,2,-1
"Basic Problem B-02",0,5,-1
"Basic Problem B-03",0,1,-1
"Challenge Problem B-10",0,4,-1
"Challenge Problem B-11",0,5,-1
"Challenge Problem B-12",0,1,-1

And

$ sed -n '/{/,$p' file
{
    "Basic Problems B": {
        "Incorrect": "0",
        "Skipped": "12",
        "Correct": "0",
        "Set": "Basic Problems B"
    },
    "Challenge Problems B": {
        "Incorrect": "0",
        "Skipped": "12",
        "Correct": "0",
        "Set": "Challenge Problems B"
    }
}

Now, you already deal with the non-JSON perfectly well, so I won't change that. Ideally, the JSON data should be parsed using a JSON parser, like jq. Sadly, I don't know enough jq to do this properly, so the best I could come up with is this, rather inelegant, solution. At least it does do what you want (replace cat file with your python submit.py --provider gt --assignment error-check command:

$ cat file | sed -n 's/[,"]//g; s/^ *//; /{/,$p'  | tac | awk -F': ' 'BEGIN{printf "%-30s%-10s%-10s%-10s\n", "Set", "Incorrect", "Skipped", "Correct"} NF==2 && !/\{/{if($1=="Set"){set=$2;data[set]["Incorrect"] = 0;data[set]["Skipped"] = 0;data[set]["Correct"] = 0;} data[set][$1]=$2}END{for(set in data){printf "%-30s%-10s%-10s%-10s\n", set,data[set]["Incorrect"],data[set]["Skipped"],data[set]["Correct"]}}' 
Set                           Incorrect Skipped   Correct   
Challenge Problems B          0         12        0         
Basic Problems B              0         12        0

Putting all this together in a shell script gives:

#!/bin/bash

tmpFile=$(mktemp)
python submit.py --provider gt --assignment error-check > "$tmpFile";

sed '/{/,$d' "$tmpFile" | column -t -s, 
sed -n 's/[,"]//g; s/^ *//; /{/,$p' "$tmpFile" |
  tac |
  awk -F': ' '
    BEGIN{
      printf "%-30s%-10s%-10s%-10s\n", "Set", "Incorrect", "Skipped", "Correct"
    }
    NF==2 && !/\{/{
      if($1=="Set"){
         set=$2;
         data[set]["Incorrect"] = 0;
         data[set]["Skipped"] = 0;
         data[set]["Correct"] = 0;
      } 
      data[set][$1]=$2
    }
    END{
       for(set in data){
         printf "%-30s%-10s%-10s%-10s\n", set, 
                                     data[set]["Incorrect"], 
                                     data[set]["Skipped"], 
                                     data[set]["Correct"]}
    }' 
rm "$tmpFile"

Which produces the following output:

$ foo.sh
Problem                   Correct?  Correct Answer  Agent's Answer
"Challenge Problem B-04"  0         4               -1
"Basic Problem B-12"      0         1               -1
"Challenge Problem B-05"  0         6               -1
"Challenge Problem B-07"  0         6               -1
"Challenge Problem B-06"  0         3               -1
"Basic Problem B-11"      0         1               -1
"Basic Problem B-10"      0         3               -1
"Challenge Problem B-03"  0         3               -1
"Challenge Problem B-02"  0         1               -1
"Challenge Problem B-01"  0         6               -1
"Challenge Problem B-09"  0         4               -1
"Challenge Problem B-08"  0         4               -1
"Basic Problem B-08"      0         6               -1
"Basic Problem B-09"      0         5               -1
"Basic Problem B-04"      0         3               -1
"Basic Problem B-05"      0         4               -1
"Basic Problem B-06"      0         5               -1
"Basic Problem B-07"      0         6               -1
"Basic Problem B-01"      0         2               -1
"Basic Problem B-02"      0         5               -1
"Basic Problem B-03"      0         1               -1
"Challenge Problem B-10"  0         4               -1
"Challenge Problem B-11"  0         5               -1
"Challenge Problem B-12"  0         1               -1
Set                           Incorrect Skipped   Correct   
Challenge Problems B          0         12        0         
Basic Problems B              0         12        0

It feels hacky though, and I hope someone can come up with a cleaner solution with dedicated JSON parsers.

Steeldriver was nice enough to give a proper jq solution in a comment, so if we incorporate that, we get the far simpler (and safer):

#!/bin/bash

tmpFile=$(mktemp)
python submit.py --provider gt --assignment error-check > "$tmpFile";

sed '/{/,$d' "$tmpFile" | column -t -s, 
sed -n '/{/,$p' "$tmpFile" | 
  jq -r '["Set","Incorrect","Skipped","Correct"], (.[] | [.Set,.Incorrect,.Skipped,.Correct]) | @tsv'
 rm "$tmpFile"

Related Solutions

How to parse JSON with shell scripting in Linux

The availability of parsers in nearly every programming language is one of the advantages of JSON as a data-interchange format.

Rather than trying to implement a JSON parser, you are likely better off using either a tool built for JSON parsing such as jq or a general purpose script language that has a JSON library.

For example, using jq, you could pull out the ImageID from the first item of the Instances array as follows:

jq '.Instances[0].ImageId' test.json

Alternatively, to get the same information using Ruby's JSON library:

ruby -rjson -e 'j = JSON.parse(File.read("test.json")); puts j["Instances"][0]["ImageId"]'

I won't answer all of your revised questions and comments but the following is hopefully enough to get you started.

Suppose that you had a Ruby script that could read a from STDIN and output the second line in your example output[0]. That script might look something like:

#!/usr/bin/env ruby
require 'json'

data = JSON.parse(ARGF.read)
instance_id = data["Instances"][0]["InstanceId"]
name = data["Instances"][0]["Tags"].find {|t| t["Key"] == "Name" }["Value"]
owner = data["Instances"][0]["Tags"].find {|t| t["Key"] == "Owner" }["Value"]
cost_center = data["Instances"][0]["SubnetId"].split("-")[1][0..3]
puts "#{instance_id}\t#{name}\t#{cost_center}\t#{owner}"

How could you use such a script to accomplish your whole goal? Well, suppose you already had the following:

a command to list all your instances
a command to get the json above for any instance on your list and output it to STDOU

One way would be to use your shell to combine these tools:

echo -e "Instance id\tName\tcost centre\tOwner"
for instance in $(list-instances); do
    get-json-for-instance $instance | ./ugly-ruby-scriptrb
done

Now, maybe you have a single command that give you one json blob for all instances with more items in that "Instances" array. Well, if that is the case, you'll just need to modify the script a bit to iterate through the array rather than simply using the first item.

In the end, the way to solve this problem, is the way to solve many problems in Unix. Break it down into easier problems. Find or write tools to solve the easier problem. Combine those tools with your shell or other operating system features.

[0] Note that I have no idea where you get cost-center from, so I just made it up.

Openweathermap json to csv

Regular expressions and sed aren't a great match for this kind of task, but it's fairly simple using jq, a JSON processing tool:

$ jq -r '.list[]|[.dt, .main[]] | @csv' < data.json
1519333200,271.62,271.62,272.921,1028.3,1037.2,1028.3,88,-1.3
1519344000,271.22,271.22,272.193,1028.11,1037.04,1028.11,100,-0.98

jq takes in a JSON data stream and filters it to extract the data you want, analogously to sed but for structured data rather than text. This command takes each element of the "list" array (.list[]) and plucks out the "dt" field (.dt) of the object inside along with everything in the "main" object too (.main[]), putting the whole lot into an array and converting that to a CSV row (@csv).

If you're interested in particular fields, or you want to reorder them, you can list them out too:

$ jq -r '.list[]|[.dt, .main.temp_min, .main.humidity] | @csv' < data.json
1519333200,271.62,88
1519344000,271.22,100

Just list out each one you care about and ignore the rest.

If you want to get into other parts of the data, you can put that in there too: .clouds will look into the clouds field, or you can read the wind speed with .wind.speed - always a dot before the name, like you were writing JavaScript. The jq manual is quite comprehensive for whatever custom behaviour you need.

Best Answer

Related Solutions

How to parse JSON with shell scripting in Linux

Openweathermap json to csv

Related Question