带点符号的 Postgres 查询行构建一个嵌套的 json 作为查询结果

Question

I have a flat table with dot-notationed columns, similar to the following:

title.en	title.fr	category.name.en	category.name.fr	category.acronym.en	category.acronym.fr
English Title	French Title	English Category Name	French Category Name	English Category Acronym	French Category Acronym

the dot notations are there to inidicate a nested object, so each dot makes a nesting level of json. According to this, I want to be able to query this result from the table in json(b):

{
  "data": {
    "title": {
      "en": "English Title",
      "fr": "French Title"
    },
    "category": {
      "name": {
        "en": "English Category Name",
        "fr": "French Category Name"
      },
      "acronym": {
        "en": "English Category Acronym",
        "fr": "French Category Acronym"
      }
    }
  }
}

I know it's possible to manually do this using nested "jsonb_build_object" functions, but want to know if it's possible to shortcut it using dot notations in column names.

Answer 1

Let's consider that data is the table name. The solution below requires several steps:

1. Get the column names from the database schema:

SELECT table_name, column_name
  FROM information_schema.columns
 WHERE table_name = 'data'

Result:

table_name	column_name
data	title.en
data	title.fr
data	category.name.en
data	category.name.fr
data	category.acronym.en
data	category.acronym.fr

2. Convert the dot-notationed column names into the target jsonb format:

SELECT ('{"' || table_name || '": {"' || replace(column_name, '.', '": {"') || '": "' || column_name || '"' || repeat('}', regexp_count(column_name, '\.')+2)) :: jsonb
  FROM information_schema.columns
 WHERE table_name = 'data'

Result:

jsonb
{"data": {"title": {"en": "title.en"}}}
{"data": {"title": {"fr": "title.fr"}}}
{"data": {"category": {"name": {"en": "category.name.en"}}}}
{"data": {"category": {"name": {"fr": "category.name.fr"}}}}
{"data": {"category": {"acronym": {"en": "category.acronym.en"}}}}
{"data": {"category": {"acronym": {"fr": "category.acronym.fr"}}}}

3. Merge the jsonb column names all together to build the final jsonb template:

In this step, we need to create a specific aggregate function jsonb_merge_agg :

CREATE OR REPLACE FUNCTION jsonb_merge(x jsonb, y jsonb) RETURNS jsonb LANGUAGE sql AS $$
  SELECT jsonb_object_agg
           ( COALESCE(x1->>'key', y1->>'key')
           , CASE
               WHEN x1->>'key' IS NULL THEN y1->'value'
               WHEN y1->>'key' IS NULL THEN x1->'value'
               WHEN jsonb_typeof(x1->'value') = 'object' AND jsonb_typeof(y1->'value') = 'object' THEN jsonb_merge(x1->'value', y1->'value')
               ELSE jsonb_build_array(x1->'value', y1->'value')
             END
           )
    FROM jsonb_path_query(x,'$[*].keyvalue()') AS x1
    FULL OUTER JOIN jsonb_path_query(y,'$[*].keyvalue()') AS y1
      ON y1->>'key' = x1->>'key' ;
$$ ;

CREATE OR REPLACE AGGREGATE jsonb_merge_agg(x jsonb)
(stype = jsonb, sfunc = jsonb_merge) ;

SELECT jsonb_merge_agg(('{"' || table_name || '": {"' || replace(column_name, '.', '": {"') || '": "' || column_name || '"' || repeat('}', regexp_count(column_name, '\.')+2)) :: jsonb) AS jsonb_template
  FROM information_schema.columns
 WHERE table_name = 'data'

Result:

jsonb_template
{"data": {"title": {"en": "title.en", "fr": "title.fr"}, "category": {"name": {"en": "category.name.en", "fr": "category.name.fr"}, "acronym": {"en": "category.acronym.en", "fr": "category.acronym.fr"}}}}

4. Replace in the jsonb_template every column name by the corresponding value:

In this step, we need to create another specific aggregate function replace_agg based on the standard replace function:

CREATE OR REPLACE FUNCTION replace(x text, y text, old text, new text) RETURNS text LANGUAGE sql AS $$
SELECT replace(COALESCE(x, y), old, new) ; $$ ;

CREATE OR REPLACE AGGREGATE replace_agg(x text, old text, new text)
(stype = text, sfunc = replace) ;

5. Build the final query:

WITH list AS (
SELECT jsonb_merge_agg(('{"' || table_name || '": {"' || replace(column_name, '.', '": {"') || '": "' || column_name || '"' || repeat('}', regexp_count(column_name, '\.')+2)) :: jsonb) AS jsonb_template
  FROM information_schema.columns
 WHERE table_name = 'data'
)
SELECT replace_agg( l.jsonb_template :: text
                  , content->>'key'
                  , content->>'value'
                  ) AS final_result
  FROM list AS l
 CROSS JOIN data AS d
 CROSS JOIN LATERAL jsonb_path_query(to_jsonb(d), '$.keyvalue()') AS content
 GROUP BY d

Final Result:

final_result
{"data": {"title": {"en": "English Title", "fr": "French Title"}, "category": {"name": {"en": "English Category Name", "fr": "French Category Name"}, "acronym": {"en": "English Category Acronym", "fr": "French Category Acronym"}}}}

all the test results in dbfiddle