Convert pg_dump to SQLite3

Working from a pg_dump file

pg_dump has enough flags to change what its output even contains, and a few of them decide whether the file in front of you is convertible at all. These are the ones that matter here.

• →pg_dump -Fc and a plain pg_dump > file.sql both work, and so does --inserts or --column-inserts, though neither is necessary: the default COPY output is read too.
• →--schema-only gives you the tables with no rows in them. That is a valid conversion, and almost never the one that was wanted.
• →--data-only is the opposite problem: rows arrive with no CREATE TABLE to name their columns, and the conversion says so rather than inventing names for them.
• →Ownership and privilege statements are dropped while the archive is unpacked, so a dump taken as one role converts without that role having to exist anywhere.

The .sqlite3 name and what carries over

The 3 in sqlite3 is the file format version and the name of the command-line tool, not a property of your conversion. One database file arrives, and the two things worth knowing about it are what it is called and what it leaves behind.

• →The file inside the zip is always database.sqlite, whichever spelling brought you to this page. Rename it to database.sqlite3, or to whatever a framework expects, once you have it - the header is what identifies it, never the extension.
• →An auto-increment column arrives holding all of its values and nothing that assigns new ones. No AUTOINCREMENT keyword is written, and the column is declared BIGINT rather than exactly INTEGER, so it is not an alias for the rowid either: rows you insert later have to supply their own key.
• →Only the rows are transactional. The very first CREATE TABLE runs in autocommit, and everything after it - that table's rows, and every later table - sits in one transaction that commits after the last row. A conversion that fails partway therefore leaves a file holding an empty shell of the first table and nothing else. You never receive that file: the zip is only assembled once the writer has returned.

Reading pg_dump

• →Reads pg_dump custom-format archives (.backup, .dump) — pgAdmin's default backup option — as well as plain SQL dumps.
• →Data written as COPY blocks is read, not just INSERT statements. COPY is pg_dump's default, so a reader that only understands INSERT returns your tables with none of their rows.
• →Primary keys declared by a later ALTER TABLE, which is how pg_dump writes them, are preserved.

Writing SQLite3

• →Columns get real declared types rather than everything being TEXT.
• →Primary keys and NOT NULL constraints are preserved.
• →A money column is declared NUMERIC(10,2), but that declaration does not stop SQLite’s own numeric affinity from converting the inserted value into a float once it lands, so 19.99 can come back with a rounding tail. Pick the DuckDB target instead when the exact decimal has to survive.

Convert pg_dump to something else

• MySQL
• PostgreSQL
• Oracle
• SQL Server
• DuckDB
• CSV
• Excel

Convert something else to SQLite3

• CSV
• Excel
• SQL dump
• DBF
• JSON
• XML
• Parquet
• DuckDB
• MS Access