fix(python): back S3BufferedReader with a real BufferedReader (#12456)

S3BufferedReader subclassed BufferedReader without initializing it, so
every method it did not override (readline, iteration, readinto, closed,
TextIOWrapper) failed with "I/O operation on uninitialized object".
The HTTP chunks now feed a RawIOBase that BufferedReader wraps, which
gives the whole file API and replaces the hand-written read/read1/peek.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Ruben Fiszel
2026-11-16 10:30:01 +01:00
committed by GitHub
co-authored by Claude Opus 5.5
parent 2cb29f87a0
commit 9a074b1c73
2 changed files with 56 additions and 53 deletions
@@ -1,5 +1,8 @@
"""Unit tests for S3BufferedReader: no network or env needed."""
import csv
import io
from wmill.s3_reader import S3BufferedReader, bytes_generator
CHUNKS = [b"AAAAAAAAAA", b"BBBBBBBBBB", b"CCCCCCCCCC"]
@@ -95,3 +98,23 @@ def test_read1_stops_after_one_chunk():
assert reader.read1(-1) == b"AAAAAAAAAA"
assert reader.read1(4) == b"BBBB"
assert counting.pulls == 2
CSV_CHUNKS = [b"name,val", b"ue\nalice,1\nb", b"ob,2\n"]
def test_text_mode_csv_reads_rows_across_chunks():
reader = make_reader(CSV_CHUNKS)
rows = list(csv.reader(io.TextIOWrapper(reader, encoding="utf-8")))
assert rows == [["name", "value"], ["alice", "1"], ["bob", "2"]]
def test_line_and_buffer_reads_stop_where_asked():
counting = CountingIterator(CSV_CHUNKS)
reader = make_reader(counting)
assert next(iter(reader)) == b"name,value\n"
assert counting.pulls == 2
target = bytearray(4)
assert reader.readinto(target) == 4
assert target == b"alic"
assert list(reader) == [b"e,1\n", b"bob,2\n"]
+33 -53
View File
@@ -1,9 +1,33 @@
from io import BufferedReader, BytesIO
from typing import Optional, Union
from io import BufferedReader, BytesIO, RawIOBase
from typing import Iterator, Optional, Union
import httpx
class _HttpChunkStream(RawIOBase):
"""Raw stream over the HTTP response's chunks, for BufferedReader to buffer."""
def __init__(self):
self.chunks: Optional[Iterator[bytes]] = None
self._pending = memoryview(b"")
def readable(self):
return True
def readinto(self, buffer):
if self.chunks is None:
raise ValueError("S3BufferedReader must be opened with `with` before reading")
while not self._pending:
try:
self._pending = memoryview(next(self.chunks))
except StopIteration:
return 0
n = min(len(buffer), len(self._pending))
buffer[:n] = self._pending[:n]
self._pending = self._pending[n:]
return n
class S3BufferedReader(BufferedReader):
"""Streaming buffered reader for S3 files via Windmill's S3 proxy.
@@ -15,6 +39,11 @@ class S3BufferedReader(BufferedReader):
storage: Optional storage backend identifier
"""
def __init__(self, workspace: str, windmill_client: httpx.Client, file_key: str, s3_resource_path: Optional[str], storage: Optional[str]):
# Every read, line and text-mode operation comes from BufferedReader
# itself; overriding them one by one leaves the rest of the file API
# failing on an uninitialized object.
self._raw = _HttpChunkStream()
super().__init__(self._raw)
params = {
"file_key": file_key,
}
@@ -28,7 +57,6 @@ class S3BufferedReader(BufferedReader):
params=params,
timeout=None,
)
self._buffer = bytearray()
def __enter__(self):
reader = self._context_manager.__enter__()
@@ -43,59 +71,11 @@ class S3BufferedReader(BufferedReader):
request=reader.request,
response=reader
)
self._iterator = reader.iter_bytes()
self._raw.chunks = reader.iter_bytes()
return self
def peek(self, size=0):
"""Return buffered bytes without consuming them.
Reads the underlying stream at most once, so the amount returned may be
more or less than `size`.
"""
if not self._buffer:
self._fill(1)
return bytes(self._buffer)
def _fill(self, limit):
# iter_bytes() yields whole HTTP chunks (~64KB), so a caller asking for
# `limit` bytes has to accumulate until the buffer holds that many.
# A negative limit means drain the stream.
while limit < 0 or len(self._buffer) < limit:
try:
self._buffer.extend(next(self._iterator))
except StopIteration:
break
def read(self, size=-1):
# BufferedReader.read(None) is documented as equivalent to read(-1).
if size is None:
size = -1
self._fill(size)
if size < 0:
result = bytes(self._buffer)
self._buffer.clear()
return result
result = bytes(self._buffer[:size])
del self._buffer[:size]
return result
def read1(self, size=-1):
"""Return up to `size` bytes, reading the underlying stream at most once.
Unlike `read`, a negative `size` returns only what is already buffered
rather than draining the whole object.
"""
if size == 0:
return b""
if not self._buffer:
self._fill(1)
if size is None or size < 0:
size = len(self._buffer)
result = bytes(self._buffer[:size])
del self._buffer[:size]
return result
def __exit__(self, *args):
self.close()
self._context_manager.__exit__(*args)