[LucaCappelletti94]

pg_dump writes every table's data as COPY ... FROM stdin with the rows inline, so a dump carrying data stopped at its first COPY, which pg2sqlite refused because sqlparser used to flatten the payload. sqlparser now keeps the payload verbatim, and the statement becomes one multi-row INSERT, so the load stays as atomic as the COPY was.

The payload is decoded with PostgreSQL's own rules, the text format with its backslash escapes and a null string matched before they are resolved, and CSV with quoted fields, HEADER, QUOTE, ESCAPE, FORCE_NULL and FORCE_NOT_NULL. Each field becomes a string literal, which PostgreSQL reads with the same input function COPY calls, and the INSERT then takes the ordinary translation, so every column conversion applies to it as it does to a written-out insert. Options PostgreSQL rejects, a payload without its \. line and a row with the wrong number of fields are refused, as are binary, FREEZE, ENCODING and FORCE_QUOTE. COPY TO and file or program sources stay refused.

A restore runs as the table owner, whom PostgreSQL lets past row-level security, so a COPY into a secured table writes its backing table, where the write exemption decides as for any trusted write.

This stacks on #154, which makes the boolean and numeric text pg_dump writes land as PostgreSQL stores it.

[codecov]

Codecov Report

❌ Patch coverage is 98.93617% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/impls/translator_impls/copy.rs	99.14%	3 Missing ⚠️
src/impls/translator_impls/insert.rs	92.30%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!