[G1gg1L3s]

This PR implement skipping postgres packets from database after encoding error, almost in the same way as the previous PR does for mysql. It also adds a bunch of tests for different cases:

• Simple queries
• Extended queries (and prepared statements, that are basically the same under the hood).
• Batched extended queries (when database sends n packets and waits for the next request)

The only issue with postgres is that a state of transaction is stored at database side, and changes in case of some errors (even parsing ones). When the state is changed, commit is basically rollback:

test=# begin;
BEGIN
test=*# select 1/0;
ERROR:  division by zero
test=!# commit;
ROLLBACK

But our error will not trigger that:

test=# begin;
BEGIN
test=*# select value_bytes from testtype;
ERROR:  encoding error in column "value_bytes"
test=# commit;
COMMIT

So, for now I will mention it in the docs. Currently there is no straight and simple solution.

P.S. test.py crossed the milestone of 10 thousands lines, so my congratulations! It takes 5 second for my formatter to proceed the file. That's what I call success.

Checklist

• Change is covered by automated tests
• The coding guidelines are followed
• Public API has proper documentation in the Acra documentation site or has PR on documentation repository
  with new changes
• CHANGELOG.md is updated (in case of notable or breaking changes)
• CHANGELOG_DEV.md is updated
• Benchmark results are attached (if applicable)
• Example projects and code samples are up-to-date (in case of API changes)

[Lagovas]

I fiound that python has high-level helper asyncio.run for that. Lets start to use it?) Works since python3.7.

[G1gg1L3s]

I wanted to use that, because I use it in small scripts myself. But I've read from the docs:

This function cannot be called when another asyncio event loop is running in the same thread.

I don't know whether our test run in parallel, but if they do, this can cause a problem (or I just don't understand this statement correctly)

[Lagovas]

then lets create new event loop for that?) use new_event_loop instead of get_event_loop

[G1gg1L3s]

Okay, but I don't quite understand the reasons. Could you explain why?

[Lagovas]

it will help run code in parallel if need. every thread or test case will have own loop and it will safe to stop loop on failure and don't touch other tests that can rely on common loop in current thread

[Lagovas]

hm, can you explain why we expect None here? As I understand:

1. we inserted data and it is already in the database
2. We tried to get this data in the transaction, took error from acra side and handled exception
3. We try to get same data but get None as result, not exception due to same error and not error that select returns no data?

[G1gg1L3s]

You are a little wrong, here is the correct workflow:

1. We insert corrupted data that will trigger decryption error and so it is accessible for the next operations.
2. We begin transaction and insert another row of data.
3. We trigger a decryption error and expect that db-driver does a rollback
4. We check that indeed, because of the rollback, the data that we insert in the transaction, has not been saved.

[Lagovas]

oh, thanks. can you add some comment above this check with explanation why we expect here None?
Because I'm sure that someone will forget that most of ORM use ROLLBACK on any exception. Or mention that after that we expect that ORM will do ROLLBACK. Just for case if event this ORM will change his logic and stop do ROLLBACK on errors.