laurentgo opened a new issue, #1337:
URL: https://github.com/apache/arrow-java/issues/1337
`ArrowMessage.frame(BufferAllocator, InputStream)`
(`flight/flight-core/src/main/java/org/apache/arrow/flight/ArrowMessage.java`)
is a hand-written parser for `FlightData`. Its `switch (tag)` ends with:
```java
default:
// ignore unknown fields.
```
but it only consumes the tag; the field's payload is never skipped. The loop
then reads the payload bytes as the next tag. Payload bytes `0x0A`, `0x12` and
`0x1A` equal `DESCRIPTOR_TAG`, `HEADER_TAG` and `APP_METADATA_TAG`, so the
payload of an unknown field can be misread as a descriptor, an IPC header or
application metadata. Depending on how the lengths line up this raises a
confusing error or silently replaces data. A known field number arriving with
an unexpected wire type falls into the same `default` (the switch matches the
whole tag).
Protobuf requires unknown fields to be skipped. The C++ fast path does it
correctly (`WireFormatLite::SkipField` in
`cpp/src/arrow/flight/serialization_internal.cc` in apache/arrow).
apache/arrow#26885 (ARROW-10962) already aligned this parser with protobuf
semantics for *missing* fields.
Reproduction: parse, through
`ArrowMessage.createMarshaller(allocator).parse(...)`, a frame holding a valid
descriptor, header and app_metadata plus an unknown field, e.g. a varint field
4 with value 10, or a length-delimited field 5 whose payload is `0A 12 1A ...`.
The result differs from, or fails to parse compared with, the same frame
without the unknown field.
Impact: any peer sending extra fields can break Java receivers on this path,
and it blocks adding fields to `FlightData` (for example a sequence number or a
control message).
Version: main (20.0.0-SNAPSHOT), any platform.
🤖 _This issue was drafted and posted by Claude Code on Laurent's behalf._
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
