magicstack/py-pgproto

Low-level IO utilities for PosgtreSQL drivers.

py-pgproto's Issues

Unable to build on Python 3.12

Getting a lot of these errors when trying to build https://github.com/MagicStack/asyncpg on Python 3.12:

  asyncpg/pgproto/pgproto.c:40800:55: error: no member named 'ob_digit' in 'struct _longobject'
              const digit* digits = ((PyLongObject*)x)->ob_digit;
                                    ~~~~~~~~~~~~~~~~~~  ^
  asyncpg/pgproto/pgproto.c:40855:55: error: no member named 'ob_digit' in 'struct _longobject'
              const digit* digits = ((PyLongObject*)x)->ob_digit;
                                    ~~~~~~~~~~~~~~~~~~  ^
  asyncpg/pgproto/pgproto.c:41034:55: error: no member named 'ob_digit' in 'struct _longobject'
              const digit* digits = ((PyLongObject*)x)->ob_digit;
                                    ~~~~~~~~~~~~~~~~~~  ^

I believe this can be solved by converting ((PyLongObject*)x)->ob_digit to ((PyLongObject*)x)->long_value->ob_digit

Tiny fix in hton.h

Should the argument x here have type int32_t?

pack_int32(char *buf, int64_t x)

I understand that the current version will produce exactly the same result.

Some more cython optimizations ?

While working on ChunkedBuffer:

https://github.com/tzickel/chunkedbuffer

I think it's for calling python methods which are already bound to a Python object, it's faster to do instead of (check the annotated output):

self._bufs_popleft(...)

this:

with cython.optimize.unpack_method_calls(False):
    self._bufs_popleft(...)

^^- but it's better to leave it though for calling other stuff, like when calling an .find directly.

Also instead of:

self._bufs = collections.deque()

it's less code

from collections import deque
self._bufs = deque()

Wrong number of items in using codecs API

The codecs.Codec.decode() and encode() APIs return a tuple, while the current code is assuming its first item:

py-pgproto/codecs/text.pyx

Lines 17 to 18 in e7f1a02

	encoded = settings.get_text_codec().encode(obj)
	cpython.PyBytes_AsStringAndSize(encoded, cstr, size)

py-pgproto/codecs/text.pyx

Lines 42 to 43 in e7f1a02

	bytes = cpython.PyBytes_FromStringAndSize(data, len)
	return settings.get_text_codec().decode(bytes)

This leads to issues in asyncpg when client_encoding is not UTF-8 - all text results are tuples, while text arguments fail to be encoded with errors.

macaddr missed

Trying copy_records_to_table i have this error:

no binary format encoder for type macaddr (OID 829)