Input
Overview
Four ways to feed bytes, from simplest to most general. Source: src/input/.
Types
pub const ReadFn = *const fn (payload: ?*anyopaque, byte_index: u32, position: Point, bytes_read: *u32) ?[*]const u8;
pub const Input = struct { payload: ?*anyopaque = null, read: ReadFn, encoding: InputEncoding = .utf8 };
pub fn chunk(self: Input, byte_index: u32, position: Point) ?struct { ptr: [*]const u8, len: u32 }MemorySource { bytes } adapts a slice (source.input()); ReaderSource { reader: *Io.Reader, buffer } adapts a std.Io.Reader with a caller buffer. StreamBuffer is the incremental pull buffer behind Parser.parseStream: require(end) pulls chunks until [0, end) is buffered or the callback reports EOF, truncate(len) drops unread read-ahead, and takeOwned() transfers the cache (becoming Tree.source with no extra copy). Source is a union { bytes, input } with byteLen, byteAt, slice, and pointAt helpers plus sourceFromBytes / sourceFromInput constructors.
Return null or length 0 from read at end of input. Offsets are pulled monotonically. Inputs are limited to 4 GiB (u32 offsets throughout the runtime).
Encodings
Input.encoding selects .utf8 (default), .utf16_le, .utf16_be, or .custom (raw bytes, treated as UTF-8). UTF-16 inputs are transcoded via unicode.utf16.transcodeToUtf8 (also exposed as treesitter.transcodeUtf16ToUtf8): matching BOMs are stripped, conflicting BOMs (UnexpectedBOM), lone surrogates (LoneSurrogate), and truncation (Truncated) are errors — never silent replacement. Tree offsets always refer to the UTF-8 form.
Ownership
Payloads and buffers stay caller-owned and must live through the parse call. Trees copy what they need.