This started as a quick hack some years ago, when I broke my PostgreSQL database (must have been version 7.3 or so) so that pg_dump wouldn't dump it anymore. Plus it couldn't handle dumping say only functions from certain schema. I have since then learned how to fix my database and pg_dump got options like --schema.
But the idea of a database being able to dump itself more autonomously persisted. After all, if LISP can do it, why not Postgres with its awesome SQL power? Information is all there in the system catalogs. One just needs to decipher it.
This tool uses information schema and standard SQL as much as possible, but for any actual practical use decoding of PostgreSQL system catalogs is required. Querying pg_catalog can turn out to be quite complicated for a somewhat casual SQL user. This project attempts to collect many SQL snippets made over the years into a coherent extension.
This will hopefully help to keep relevant SQL code for these things in one place. Perhaps this can also be of some use to GUI tools builders and such in implementing "Extract DDL" functionality.
Support for other postgres objects: - ✔︎ pg_amop, pg_amproc - ✔︎ extensions - pg_default_acl - pg_largeobject_metadata - pg_statistic_ext - pg_publication_rel - pg_enum
Support for other missing options: - ✔︎ comments on all objects - ✔︎ ownership of all objects - ✔︎ grants on all objects - grants vs current_role (who is grantor?) - materialized view tablespace & storage parameters - table of type not nulls - PG14 SET COMPRESSION - SET STATISTICS on indexes - ✔︎ pg_subscription not readable by non superuser
- figure out how to elegantly separate pre-data, post-data, create, alter and dcl
- move not nulls to constraints section
- move storage setting to pre-data section
- use ONLY when appropriate
- ✔︎ add
- group column alters together by column name
- improve dumping of comments (be quiet on NULL comments)
Currently NULL comments are included, to encourage commenting.
- ✔︎ mysterious duplicates in index section for partitioned tables (see table dept_1)
- perhaps include table subpartitions in a script
- handle dependancies for types better (use shell types)
- handle sequences better (create if not exists)
- ✔︎ do not emit ALTER OWNER for objects owned by current role
- improve and add to simple tests
- improve support for non superusers (more testing, etc)
- make some tests to test if what we output actually runs, test execute them
- make some tests which compare to output of pg_dump;
make utility for any sql file to compare the dump by pg_dump and ddlx.
Comparison should compare actual contents, not merely text.
Some options as to what and how to dump stuff might be required:
DROP- generate DROP statements
ALTER- prefer ALTER to CREATE, implies 'INE' and 'IE'
DCL- include DCL (GRANTS)
COR- use CREATE OR REPLACE where possible
INE- use IF NOT EXISTS where possible
IE- use IF EXISTS where possible
WRAP- wrap in BEGIN / END
EXT- include objects from extensions. Normally, these are omitted.
DEP- output objects which depend on this object too
DATA- add statements preserve / copy table data
These might be passed as optional second arg to extractor functions Perhaps as a text array? JSON?
Also, consider how to choose pre and post data DDL.
Perhaps there are other ways to implement some of this?
Other DDL dumping tools
Command line client
psql contains lots of packaged SQL for handling metadata
mainly to support code completion and various \d* commands.
Database dump tool
pg_dump contains lots of packaged SQL for handling metadata
and especially considers various dependancies. Source code for main C source file
is is about 20000 lines long.
PgAdmin3 DDL generation and schema handling code is an interesting mix of wxWidgets GUI toolkit (C++) and SQL. It requires GUI. Generated DDLs are database administrator friendly.
PgAdmin4 is the stuff of legends