v1.8.0

Added

• Public open() method on AsyncGzipBinaryFile and AsyncGzipTextFile for the explicit try/finally lifecycle; __aenter__ now delegates to it. Calling open() on an already-open file raises ValueError, and a closed instance cannot be reopened (matching io objects). A failed open() (e.g. a transient write error on an external fileobj) leaves the instance retryable instead of half-open.
• __repr__ on both file classes showing the name, mode, and closed state. The repr is safe even on partially-constructed instances, so debuggers and locals-capturing traceback formatters get a usable string instead of AttributeError.
• Typing overloads on the AsyncGzipFile factory so the return type narrows to AsyncGzipBinaryFile or AsyncGzipTextFile from the mode literal.

Fixed

• Text-mode forward seek() no longer corrupts decoder state when bytes were consumed directly from the public buffer accessor: the fast path now tracks the decoder's byte frontier and falls back to a full replay when the binary position has moved past it (previously UnicodeDecodeError, or silent corruption with error-tolerant encodings).
• readinto()/readinto1() accept any writable buffer, including those with itemsize > 1 such as array.array, filling them and returning byte counts exactly like stdlib gzip.
• readinto() fills its internal buffer before consuming, so a decompression error mid-request leaves the stream position and already-decoded data intact for salvage, matching read().
• readinto1() and read1() repeat fills until at least one byte decodes, so a 0/b"" result now means EOF (a single fill can consume compressed input, e.g. the gzip header, without producing output); while await f.readinto1(buf): ... loops are safe.
• peek() on a closed file raises ValueError("I/O operation on closed file.") like the other read methods, and readinto1()'s writable-buffer TypeError names the right method.

Performance

• readinto()/readinto1() write decompressed data straight into the caller's buffer instead of allocating an intermediate bytes object.
• Forward plain-offset text seeks replay only the delta from the current position instead of restarting decompression from byte 0 — including when decoded read-ahead is buffered, since the buffer is discarded by the seek anyway.
• Universal-newline detection counts terminators with windowed str.count instead of replace()-based scans, eliminating two chunk-sized string allocations per chunk on CRLF streams.

Documentation

• Documented that text tell() cookies are bound to the handle that minted them, and added the resumable-processing recipe: checkpoint plain decompressed-byte offsets at line boundaries via the binary layer, then resume by seeking in text mode. docs/api.md summarizes and links to the canonical recipe in docs/index.md.

Testing

• Hypothesis property-based parity suite against stdlib gzip: randomized multi-member archives are read through both libraries in lockstep across access patterns, and single-byte corruption must be detected identically — aiogzip may be stricter than stdlib only for reserved FLG-bit flips, which zlib rejects and stdlib's header parser ignores.