Dmytro Shteflyuk created THRIFT-6105:
----------------------------------------
Summary: Ruby native MemoryBufferTransport should retain partial
read progress
Key: THRIFT-6105
URL: https://issues.apache.org/jira/browse/THRIFT-6105
Project: Thrift
Issue Type: Bug
Components: Ruby - Library
Reporter: Dmytro Shteflyuk
Assignee: Dmytro Shteflyuk
h3. Problem
The native {{MemoryBufferTransport#read_into_buffer}} implementation keeps the
read position in a local variable until the entire operation completes. If it
raises after copying one or more bytes, the destination has been modified but
the transport still reports those copied bytes as available. The pure-Ruby
implementation records each successfully copied byte before a later failure.
h3. Client impact
Clients that recover from {{IndexError}} or {{EOFError}} can receive input a
second time even though it was already copied into their destination buffer.
The observable behavior also depends on whether {{thrift_native}} is loaded.
h3. Reproduction
{code:ruby}
transport = Thrift::MemoryBufferTransport.new("ab")
destination = +"x"
begin
transport.read_into_buffer(destination, 2)
rescue IndexError
end
p [destination, transport.available]
# Pure Ruby: ["a", 1]
# Native extension before this change: ["a", 2]
{code}
The same mismatch occurs when input ends after a successful partial copy.
h3. Expected behavior
Once a byte has been copied to the destination, both implementations should
treat it as consumed even if a later destination-boundary or input-availability
check raises an exception.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
