v1.4.0 introduces breaking changes to the CryptoContext API

[j2kun]

The const qualifier is removed from Eval* methods that take plaintexts, which causes previously-working code like this to err:

  const auto& pt = cc->MakePackedPlaintext(...);
  const auto& ct1 = cc->EvalSub(pt, ct);

error: no matching member function for call to 'EvalAdd'
src/pke/include/cryptocontext.h:1485:25: note: candidate function not viable: 2nd argument ('const Plaintext' (aka 'const shared_ptr<PlaintextImpl>')) would lose const qualifier
 1485 |     Ciphertext<Element> EvalAdd(ConstCiphertext<Element>& ciphertext, Plaintext& plaintext) const {

Has the semantics of these API's changed? (Do they now mutate the plaintext?) If not, it should remain const, and if so the backwards incompatibility should be announced as part of the release notes.

[j2kun]

Also removing const for some CryptoContext methods in the past, I've noticed that the result is an explicit copy, which I'd like to avoid.

[pascoec]

Hi Jeremy, yes, the API for specific Plaintext-related functions has changed to reflect the fact that previously const-declared plaintext variables were and still are being mutated. The affected functions call SetFormat on the plaintext, which modifies some plaintext data members. The modifications were previously hidden because SetFormat was declared const and the affected data members were declared mutable. Now they are not.

  auto& pt = cc->MakePackedPlaintext(...);
  const auto& ct1 = cc->EvalSub(pt, ct);

should produce the same internal behavior from before, with pt after execution of EvalSub in the same internal state. Internally, there are no additional copies.

I agree backward incompatibility should be announced. We will add something to the release notes.