catj

#!/bin/bash
#
# Outputs JSON paths.
#
# Based on https://news.ycombinator.com/item?id=20264654
#
# Usage:
# curl -s https://raw.githubusercontent.com/sitepoint-editors/json-examples/master/src/db.json | ./jsonpaths

jq -r --stream '
  select(length > 1)
  | (
    .[0] | map(
      if type == "number"
      then "[" + tostring + "]"
      else "." + "\"" + . + "\""
      end
    ) | add
  ) + " = " + (.[1] | @json)
' < "${1:-/dev/stdin}"

uncatj

#!/bin/bash
#
# Converts flat json such as:
#
#   .clients[0].id = "59761c23b30d971669fb42ff"
#   .clients[0].isActive = true
#   .clients[0].age = 36
#
# back to json.
# The newline in the middle of the sed command is for BSD compatibility
#
# Still one problem:
#   - top-level arrays flatten to "[0].asdf" which cause invalid path expressions
#     The only hack I can think of right now for this is to wrap top level arrays 
#     as a value in a pair in the top level object:
#         sed -E 's/(^\[[0-9]\])/.TLA\1/g'
#     along with the <<< '{"TLA":[]}' listed below instead of <<< '{}'
#
#     Doing so would produce the following:
#  % echo '["g", 3, "throw"]' | catj | sed -E 's/(^\[[0-9]\])/.TLA\1/g' | uncatj
#    {
#      "TLA": [
#        "g",
#        3,
#        "throw"
#      ]
#    }
#    This ugly change also forces an extra element into normal objects:
#
#  % echo '.HostConfig.PortBindings.HostIp = ""' | uncatj
#  {
#    "TLA": [],
#    "HostConfig": {
#      "PortBindings": {
#        "HostIp": ""
#      }
#    }
#  }
# I hate it but it works

( jq -r "$(sed -E 's/(^\[[0-9]\])/.TLA\1/g ; s/$/ |/;$a\
.')" <<< '{"TLA":[]}' ) < "${1:-/dev/stdin}"