Re: nested xml/json to table

[Thomas Kellerer <shammat@xxxxxxx>]

Wim Bertels schrieb am 17.03.2023 um 11:05:
> what would be the general idea: "easily" convert an hierarchical
> structure like json or xml to a table; for example creating columns by
> appending the key-names when going doing down the three, using null for
> empty values, adding more columns as needed by the given structure.
> (1-way operation)
>
> a few conceptual gists:
> jsonX=
> {
>     "glossary": {
>         "title": "example glossary",
> 		"GlossDiv": {
>             "title": "S",
> 			"GlossList": {
>                 "GlossEntry": {
>                     			"ID": "SGML",
> 					"SortAs": "SGML",
> 					"GlossTerm": "Sta..";
> 					"Acronym": "SGML",
> 					"Abbrev": "ISO 8879:1986",
> 					"GlossDef": {
> 			                        "para": "A meta-m..",
> 						"GlossSeeAlso": 
> 							["GML", "XML"]
>                     			},
> 					"GlossSee": "markup"
>                 }
>             }
>         }
>     }
> }
>
> select *
> from json_to_table('jsonX');
> -- generated columns with no data/only nulls could be removed..
> -- arrays could be unnested in the process as well
>
> glossary | glossary.title   | glossary.title.GlossDiv.title | ..
> -------------------------------------------------------------
> null	 | example glossary | S			            |
> ..
>
> the last column: glossary.title.GlossDiv.GlossList.GlossEntry.GlossSee
> with value "markup"
>
> ---
>
> what if there are different structures that need to be combined?
> (they could be added in the same manner as before)
>
> jsonY=
> {
> s1:[{
>     "f1": "a",
>     "f2": "b",
>     "f3": { "f3.1": "c",
> 	    "f3.2": "d"}
>    },
>    {
>     "f1": "e",
>     "f4": "g"
>    }
>    ]
> }
>
> select *
> from json_to_table('jsonY');
> -- generated columns with no data/only nulls could be removed..
> -- separator sign is untrusted
>
> s1  | s1.f1 | s1.f2 | s1.f3 | s1.f3.f3.1 | s1.f3.f3.2 | s1.f4
> -------------------------------------------------------------
> null| a     | b     | null  | c          | d          | null
> null| e     | null  | null  | null       | null       | g


You can't have a function that returns a different set of columns each time you call it
(without specifying the output columns - which you don't want).

I have once written a function to flatten a JSON hierarchy to multiple rows.

Applied to your first example it would return the following:

path                                             | key          | value
-------------------------------------------------+--------------+-----------------
/glossary                                        | title        | example glossary
/glossary/GlossDiv                               | title        | S
/glossary/GlossDiv/GlossList/GlossEntry          | ID           | SGML
/glossary/GlossDiv/GlossList/GlossEntry          | Abbrev       | ISO 8879:1986
/glossary/GlossDiv/GlossList/GlossEntry          | SortAs       | SGML
/glossary/GlossDiv/GlossList/GlossEntry          | Acronym      | SGML
/glossary/GlossDiv/GlossList/GlossEntry          | GlossSee     | markup
/glossary/GlossDiv/GlossList/GlossEntry          | GlossTerm    | Sta..
/glossary/GlossDiv/GlossList/GlossEntry/GlossDef | para         | A meta-m..
/glossary/GlossDiv/GlossList/GlossEntry/GlossDef | GlossSeeAlso | ["GML", "XML"]

And the following for the second example:

path   | key  | value
-------+------+------
/s1    | f1   | a
/s1    | f2   | b
/s1/f3 | f3.1 | c
/s1/f3 | f3.2 | d
/s1    | f1   | e
/s1    | f4   | g


Thomas




create or replace function flatten(p_input jsonb, p_path text)
  returns table(path text, key text, value text)
as
$$
begin
  if jsonb_typeof(p_input) = 'array' then
    return query
      select f.*
      from jsonb_array_elements(p_input) as a(element)
        cross join flatten(a.element, p_path) f;
  else
    return query
      select p_path, e.key, e.value #>> '{}'
      from jsonb_each(p_input) as e(key, value)
      where jsonb_typeof(e.value) not in ('object', 'array')
        or (jsonb_typeof(e.value) = 'array' and jsonb_typeof(e.value -> 0) <> 'object')

      union all

      select f.*
      from jsonb_each(p_input) as t(key,value)
        cross join flatten(t.value, p_path||'/'||t.key) as f
      where jsonb_typeof(t.value) = 'object'

      union all

      select f.*
      from jsonb_each(p_input) as t(key,value)
        cross join jsonb_array_elements(t.value) as a(element)
        cross join flatten(a.element, p_path||'/'||t.key) as f
      where jsonb_typeof(t.value) = 'array'
        and jsonb_typeof(t.value -> 0) = 'object';
  end if;
end;
$$
language plpgsql
immutable
parallel safe
;

create or replace function flatten(p_input jsonb)
  returns table(path text, key text, value text)
as
$$
  select *
  from flatten(p_input, '');
$$
language sql
immutable
parallel safe
;