tree 8098d1c1f8a89c20a34ca6101db79ae9b2b8e04e
parent 453bda8fea991fd1a333fa5326b1afbcfeaafe5d
author bnc <bnc@google.com> 1632853448 -0700
committer Copybara-Service <copybara-worker@google.com> 1632853490 -0700

Simplify HttpDecoder state transitions.

Both before and after this CL, non-buffered frames go through the following
states:
  STATE_READING_FRAME_LENGTH
  STATE_READING_FRAME_TYPE
  STATE_READING_FRAME_PAYLOAD
  STATE_FINISH_PARSING
They skip STATE_READING_FRAME_PAYLOAD if the payload is empty.

Before this CL, buffered frames normally skip STATE_FINISH_PARSING, except they
skip STATE_READING_FRAME_PAYLOAD if payload is empty just like non-buffered
states, and therefore they have to go through STATE_FINISH_PARSING.
ReadFramePayload() corresponds to STATE_READING_FRAME_PAYLOAD and
FinishParsing() corresponds to STATE_FINISH_PARSING, and states are only changed
in methods that correspond to states, except for the helper
BufferOrParsePayload() which does not have a corresponding state but can still
change the state, which its caller must detect.

After this CL, the states STATE_READING_FRAME_PAYLOAD and STATE_FINISH_PARSING
only process non-buffered frames, and buffered frames go through a dedicated new
state STATE_BUFFER_OR_PARSE_PAYLOAD.  The state transition path forks in
ReadFrameLength().  BufferOrParsePayload() gets promoted from a random helper
called from other methods to a method corresponding to a state, called from
ProcessInput() which drives the state machine.  State transitions are now
cleaner.

PiperOrigin-RevId: 399487182