If `path_or_url` starts with http://, https://, or s3://, fetch it to a system temp file and return the temp path (with the original extension preserv

Add `delta` to the count for `k`. Returns true iff this added a new key.

Decompress a page body. Parquet codecs we handle: UNCOMPRESSED (0), SNAPPY (1), GZIP (2), ZSTD (6). LZO (3) / BROTLI (4) / LZ4 (5) / LZ4_RAW (7) are n

Returns true if field present; sets fid, tid. tid==0 marks end of struct.

Resolve an English noun to a column on a specific table. Synonyms are tried in order: exact, lowercase, then the synonym class the noun belongs to (if

Clean up test file.

Detect SELECT * style identity projection: every output column is a plain BoundColumnRef pointing at the same input slot. When this holds, GetData bec

Column pruning: tell the scan which columns are needed.

Write a full file replacing its previous contents. Used by LIVE VIEW tests - the mtime change is what the cache check observes.

LIKE '%needle%' substring search. memchr-anchored memcmp confirm — uses the CRT's SIMD memchr (MSVC) for the hot byte-scan, then memcmp to verify. Mic

Read one page from the given file offset. Returns bytes consumed. `body_ptr` is set to point at the (compressed) page body inside `file_base` - no cop

Fast NDJSON parser: single pass over the mmap'd/loaded buffer, no per-line std::unordered_map, no per-char std::string append. Detects schema from the

Parse only the schema header - magic bytes + column list - so callers can register the catalog entry without reading every row.

Normalize an identifier for fuzzy matching: lowercase + map `_`, ` `, and `-` to a single separator. This lets us recognise `customer_id` as referring

Parse a double from [start, end). Apple's libc++ doesn't implement `std::from_chars` for floating point, so we can't use it portably - fall back to st

Expose internal buffer for parallel processing.

Bulk-extract up to n indices from whichever run is currently active. If the active run is bit-packed, peel values in a tight loop without re-checking

Stream-parse the Arrow data directly into typed DataChunk Vectors. Each chunk holds up to VECTOR_SIZE rows. No Value-boxed rows_ stored.

List/set header: (count << 4) | elem_type, count==15 => varint count.

Resolve an English noun to a table in the schema. Tries exact match, case-insensitive, and singular<->plural. Returns nullptr if nothing matches or mu