Fix IsolationLevel.read_comitted and introduce IsolationLevel.default

[Pliner]

PR introduces IsolationLevel.default and fixes IsolationLevel.read_comitted as it was mentioned by @dvarrazzo.

Are there changes in behavior for the user?

Instead of hacking

        with (await pool.cursor()) as cur:
            # Set the connection in read committed isolation level
            # to avoid serialization failures on file lock
            # https://github.com/aio-libs/aiopg/issues/699
            # You can unlock the aiopg version from setup.py when this hack
            # is no more needed.
            cur._transaction = aiopg.Transaction(
                cur, aiopg.IsolationLevel.read_committed
            )
            yield cur

PR allows to pass IsolationLevel

        with (await pool.cursor(isolation_level=aiopg.IsolationLevel.read_committed)) as cur:
            yield cur

Related issue number

Fixes #699

Checklist

• I think the code is well written
• Unit tests for the changes exist
• Documentation reflects the changes

[codecov]

Codecov Report

Merging #770 (a65e032) into master (52532fb) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #770      +/-   ##
==========================================
+ Coverage   93.08%   93.11%   +0.03%     
==========================================
  Files          13       13              
  Lines        1532     1539       +7     
  Branches      172      172              
==========================================
+ Hits         1426     1433       +7     
  Misses         78       78              
  Partials       28       28              

Impacted Files	Coverage Δ
aiopg/connection.py	91.86% <100.00%> (ø)
aiopg/cursor.py	100.00% <100.00%> (ø)
aiopg/transaction.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 52532fb...a65e032. Read the comment docs.

[dvarrazzo]

Why readonly and deferrable are not supported? As this object is running BEGIN it should be able to support it, they are independent from the isolation level.

https://www.postgresql.org/docs/current/sql-begin.html

[dvarrazzo]

I think it was right that __slots__ was on the base class and that these properties are available to all the hierarchy

[Pliner]

I've only refactored current implementation 🙈

[Pliner]

Why readonly and deferrable are not supported? As this object is running BEGIN it should be able to support it, they are independent from the isolation level.

I dunno why it was unsupported, let me check

[Pliner]

https://www.postgresql.org/docs/13/sql-set-transaction.html

It looks like readonly supported for any isolation level, but deferred - only for serialisable with readonly.

But I'm a but unsure about current behaviour with raise of an exception.

[dvarrazzo]

You shouldn't impose a rule on top of what postgres implements.

If it's a no-op it may be a no-op only for the current implementation and change in the future. If the database accepts it, it is not the driver's business to decide to raise an exception.

This seems to work: the database doesn't raise exceptions or even warnings:

piro=# begin isolation level read uncommitted deferrable;
BEGIN
*piro=#

so I don't see a reason to special-case it and raise an exception.

[Pliner]

Agree, thanks 🎉