API Changes (Backward Incompatible)
- Support for Python 3.6 has been removed.
- Support for Python 3.7 has been removed.
- Support for Python 3.8 has been removed.
API Changes (Backward Compatible)
- Support for Python 3.9 has been added.
- Support for Python 3.10 has been added.
- Support for Python 3.11 has been added.
- Support for Python 3.12 has been added.
- Support for Python 3.13 has been added.
Bugfixes
API Changes (Backward Incompatible)
- Support for Python 2.7 has been removed.
- Support for Python 3.4 has been removed.
- Support for Python 3.5 has been removed.
- Support for PyPy (Python 2.7 compatible) has been removed.
- Support for Python 3.8 has been added.
Bugfixes
- Performance improvement of static header search. Use dict search instead of linear search.
- Fix debug output of headers during encoding.
API Changes (Backward Incompatible)
- Removed nghttp2 support. This support had rotted and was essentially non-functional, so it has now been removed until someone has time to re-add the support in a functional form.
- Attempts by the encoder to exceed the maximum allowed header table size via dynamic table size updates (or the absence thereof) are now forbidden.
API Changes (Backward Compatible)
- Added a new
InvalidTableSizeError
thrown when the encoder does not respect the maximum table size set by the user. - Added a
Decoder.max_allowed_table_size
field that sets the maximum allowed size of the decoder header table. See the documentation for an indication of how this should be used.
Bugfixes
- Up to 25% performance improvement decoding HPACK-packed integers, depending on the platform.
- HPACK now tolerates receiving multiple header table size changes in sequence, rather than only one.
- HPACK now forbids header table size changes anywhere but first in a header block, as required by RFC 7541 § 4.2.
- Other miscellaneous performance improvements.
Security Fixes
CVE-2016-6581: HPACK Bomb. This release now enforces a maximum value of the decompressed size of the header list. This is to avoid the so-called "HPACK Bomb" vulnerability, which is caused when a malicious peer sends a compressed HPACK body that decompresses to a gigantic header list size.
This also adds a
OversizedHeaderListError
, which is thrown by thedecode
method if the maximum header list size is being violated. This places the HPACK decoder into a broken state: it must not be used after this exception is thrown.This also adds a
max_header_list_size
to theDecoder
object. This controls the maximum allowable decompressed size of the header list. By default this is set to 64kB.
API Changes (Backward Compatible)
- Added
HeaderTuple
andNeverIndexedHeaderTuple
classes that signal whether a given header field may ever be indexed in HTTP/2 header compression. - Changed
Decoder.decode()
to return the newly addedHeaderTuple
class and subclass. These objects behave like two-tuples, so this change does not break working code.
Bugfixes
- Improve Huffman decoding speed by 4x using an approach borrowed from nghttp2.
- Improve HPACK decoding speed by 10% by caching header table sizes.
Bugfixes
- When passing a dictionary or dictionary subclass to
Encoder.encode
, HPACK now ensures that HTTP/2 special headers (headers whose names begin with:
characters) appear first in the header block.
API Changes (Backward Compatible)
- Added new
InvalidTableIndex
exception, a subclass ofHPACKDecodingError
. - Instead of throwing
IndexError
when encountering invalid encoded integers HPACK now throwsHPACKDecodingError
. - Instead of throwing
UnicodeDecodeError
when encountering headers that are not UTF-8 encoded, HPACK now throwsHPACKDecodingError
. - Instead of throwing
IndexError
when encountering invalid table offsets, HPACK now throwsInvalidTableIndex
. - Added
raw
flag todecode
, allowingdecode
to return bytes instead of attempting to decode the headers as UTF-8.
Bugfixes
memoryview
objects are now used when decoding HPACK, improving the performance by avoiding unnecessary data copies.
- Fixed a bug where the Python HPACK implementation would only emit header table size changes for the total change between one header block and another, rather than for the entire sequence of changes.
- Remove unused
HPACKEncodingError
. - Add the shortcut ability to import the public API (
Encoder
,Decoder
,HPACKError
,HPACKDecodingError
) directly, rather than fromhpack.hpack
.
- Add support for emitting 'never indexed' header fields, by using an optional third element in the header tuple. With thanks to @jimcarreer!
- Header fields that have names matching header table entries are now added to the header table. This improves compression efficiency at the cost of slightly more table operations. With thanks to Tatsuhiro Tsujikawa.
- Initial fork of the code from hyper.