websocket, fix curl_ws_recv()

[icing]

• when data arrived in several chunks, the collection into the passed buffer always started at offset 0, overwriting the data already there.

[calvin2021y]

maybe a unit tests can prevent this kind error.

[icing]

maybe a unit tests can prevent this kind error.

It was triggered in pytest test_20_06, but it depended on how data arrived from the server. Any suggestion for a reliable test appreciated.

[calvin2021y]

maybe use netcat and pv to limit the send speed of a exists tcp stream

nc -l -p 8888 | pv -L 512  | nc example.com 80

in this example pv limit the stream speed at 512 byte per second.

[jay]

It was triggered in pytest test_20_06, but it depended on how data arrived from the server.

Would CURL_DBG_SOCK_RMAX work for that

[icing]

It was triggered in pytest test_20_06, but it depended on how data arrived from the server.

Would CURL_DBG_SOCK_RMAX work for that

Added a new debug environment var CURL_WS_CHUNK_SIZE to influence the buffer dimensions used in en-/decoding for websockets. This exposes the bug in test_20-07 and verifies the fix.

[bagder]

Thanks!