Strings sample

This ports pgrx’s strings example to ordinary C# functions. It preserves exact text, UTF-8 byte offsets and the separator rules used by Rust’s split_terminator.

SELECT return_static();                     -- This is a static string xxx
SELECT append('hi', ' there');              -- hi therex
SELECT to_lowercase('İ ΟΣ');                -- i + combining dot, then ος
SELECT public.substring('Aé中😀Z', 6, 10);         -- 😀
SELECT split(',a,,b,', ',');                -- {"",a,"",b}
SELECT * FROM split_set(',a,,b,', ',');
SELECT i, s FROM split_table(',a,,b,', ',');

The example assumes installation in public. Use your extension’s schema when qualifying substring: PostgreSQL’s built-in function has the same name and signature but uses character positions and a length rather than byte bounds. The table form preserves the upstream i and s column names and zero-based ordinals. Sets stream pieces in source order; use the ordinal column with ORDER BY i when a query needs explicit ordering.

Lowercase uses full Unicode 17.0 mappings, including dotted-I expansion, contextual final sigma and supplementary scripts. It is independent of thread culture, database collation and the platform’s globalization provider. The data matches the reference Rust 1.97.1 standard library. It is not PostgreSQL’s collation-dependent lower function.

substring takes start-inclusive and end-exclusive UTF-8 byte offsets. Both offsets must be scalar boundaries, including empty slices. Negative, reversed, out-of-range or mid-scalar offsets are errors. The borrowed native bytes are consumed before returning an independent managed string. These ordinary C# failures report SQLSTATE 38000; pgrx’s Rust slice panics report XX000. Both fail the statement and require transaction recovery.

Splitting preserves leading and interior empty pieces and removes only the final terminator suffix. An empty separator produces a leading empty piece and then one piece per Unicode scalar. Empty input with an empty separator therefore produces one empty piece; empty input with a nonempty separator produces none. Matching is ordinal, without case folding or Unicode normalization.

All arguments are non-nullable, so Ankus declares the functions STRICT: a NULL argument returns NULL without calling them. SQL NULL is distinct from present empty text, an empty array and an empty string row. Set/table iterators retain managed text rather than borrowed spans across native callbacks.

The implementation is MIT-licensed. The sample’s third-party notice preserves the upstream pgrx license and the Unicode data notice.