The current version of 4880bis has a chunk size parameter for AEAD:
## AEAD Encrypted Data Packet (Tag 20)
The body of this packet consists of:
* A one-octet chunk size.
The chunk size octet specifies the size of chunks using the
following formula (in C), where c is the chunk size octet:
chunk_size = ((uint64_t)1 << (c + 6))
An implementation MUST support chunk size octets with values from 0
to 56. Chunk size octets with other values are reserved for future
extensions. Implementations SHOULD NOT create data with a chunk
size octet value larger than 21 (128 MiB chunks) to facilitate
buffering of not yet authenticated plaintext.
In other words, chunks can be up to 1 << (56 + 6) = 2^62 bytes large
According to RFC 5116, an AEAD algorithm must not output partially
particular implementation of an AEAD algorithm is requested to
process an input that is outside the range of admissible lengths, or
an input that is outside the range of lengths supported by that
implementation, it MUST return an error code and it MUST NOT output
any other information. In particular, partially encrypted or
partially decrypted data MUST NOT be returned.
Because it is hard to know the context (i.e., the available resources)
in which a message will be decrypted, it is difficult for an
implementation to make a reasonable choice when doing the encryption.
Actually, I'm not aware of any advantages for chunk sizes that are
larger than a few kilobytes.
Consequently, I propose not only imposing a reasonable ceiling on the
chunk size that even small embedded devices with a cortex M0 could
handle, but to simply fix the parameter to 16 KiB. It's not clear to
me that a variable size offers any advantages. But, there is a clear
disadvantage: it's unjustified complexity, which is a breeding ground
for bugs. Unless someone can justify this added complexity, I see no
reason to parameterize this.
(If it is needed, the size could be a function of the actually AEAD
algorithm, e.g., EAX, OCB, etc.)
I've attached a patch that makes the proposed change.
I've spoken with several people including Vincent Breitmoser (Open
Keychain), Justus Winter & Kai Michaelis (also Sequoia), Phil
Zimmermann, and Hanno Böck and they all support this proposal.