Strip invisible UTF-8 characters that sneak into pasted input and break parsing silently: U+200B..U+200F ZWSP / ZWNJ / ZWJ / LRM / RLM U+202A..U+202E

level may actually range from [0,10] (10 is a "hidden" max level, where we want a bit more compression and it's fine if throughput to fall off a cliff

Strip directory + extension from a file path for the fallback CTAS name. "./data/sales.csv" -> "sales".

Get cardinality (number of unique values).

Map a Parquet CompressionCodec ID to a human-readable name for error messages. Mirrors the Parquet Thrift enum.

Cast a numeric Vector to `target_type` if its current type doesn't match. Returns the original Vector by-move if no cast is needed; otherwise builds a

Count newlines in [start, end) using SWAR.

Days from civil 1970-01-01 to civil (y,m,d). Inverse of DaysToYMD above. Algorithm by Howard Hinnant; valid for any proleptic Gregorian date.

Convert days-since-1970-01-01 to (Y, M, D) using Howard Hinnant's civil-from-days. Used to render DATE values as ISO-8601 strings instead of opaque in

Decode PLAIN-encoded values from a buffer into `out`.

Evaluate predicates against a DataChunk row (for CSV / JSON sources). Returns false if any predicate fails or the row is null in a predicate col.

Local copy of physical_planner.cpp::FindSubstr — same memchr-anchored memcmp loop, kept here to avoid pulling the planner header into this side TU. ~6

Accessors.

Flush the active table: append every (key,count) to the thread's flat spill buffers (1-way — no scatter), then reset the table and arena.

File scan support: stores file path for streaming reads.

Live view support: view's storage is cached; expansion in Connection::Query checks the watched file's mtime and only re-executes when it changes. Chea

Number of worker threads.

Wrap an identifier in double quotes if it contains any character that would need quoting in SQL (spaces, punctuation, starts with a digit, or is a res

Read + decompress a dictionary page into `dict`. For VARCHAR, the dict bytes are copied into `heap` (so dict pointers remain stable across the RG). Se

Read int64 from any integer column (INTEGER or BIGINT); 0 if other.

Render `WHERE EXTRACT(YEAR FROM <col>) = <year>` with fallback to string-prefix comparison if the column looks like a VARCHAR. This is the least-surpr

Stream the rows of `table_name` into typed DataChunk vectors. Reuses the existing B-tree scan (which returns Value rows) and copies the results into t

Adopt another vector's auxiliary buffer so string_t pointers copied from the other vector remain valid. Used by vectorized operators that copy string_

Optional cached ParquetReader shared across queries - the schema- detection path in Connection::Query opens one to populate the catalog, and we stash

Used by the binder when promoting a literal to match the type of the column it's compared with (avoids stod-per-row in the executor).

Per-column hint: lengths-only mode. The decoder fills str_lengths (one uint32 per row) and skips byte materialization entirely. Use when the column is

Decompress a Snappy stream directly into a pre-allocated raw buffer. Caller pre-allocates `out` with at least the uncompressed-page-size bytes (read f

Slice the schema JSON into one substring per field. Each substring covers from the field's "name" key to the matching closing brace of the field recor