Unit tests are failing with a unicode error there in tests/test_pgexecute.py. Are you able to run them locally (running py.test in pgcli directory)?

I think that the reason of the failure is that the new get_new_connection() method should install the same typecasters as done by connect()... Maybe some refactor could be made, to avoid duplicating basically the same code to create the connection.

Also, I'm not sure if those typecasters should be moved only onto the new user_conn: I would expect that only user queries will involve JSON values...

Finally, in the method execute_normal_sql() (the only place where the new user_conn gets used), there's a loop that pops possible notices: shouldn't it targeting the same user_conn connection?

Codecov Report

Merging #917 into master will decrease coverage by 0.51%.
The diff coverage is 59.09%.

@@            Coverage Diff             @@
##           master     #917      +/-   ##
==========================================
- Coverage   84.57%   84.05%   -0.52%     
==========================================
  Files          21       21              
  Lines        2463     2496      +33     
==========================================
+ Hits         2083     2098      +15     
- Misses        380      398      +18     

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 0f0be9d...59f0c77. Read the comment docs.

Thank you for having a look at this PR.
I fixed the unicode error (due to typecasters) and notices, thank you.
It definitely need some refactor, in order to init all connections in the same place. I can have a look in the next days. This could be done in another PR, up to you.
[Edit] Made the refactor, any feedback welcome. Idk why pep8radius doesn't produce any output locally...

@gma2th I looked through the PR and I have 2 questions. Let me preface these comments by saying, I would like to minimize the number of configuration options in pgcli. It is wise to choose sensible defaults wherever possible.

1. What is the purpose of surfacing the autocommit mode as a config? Is there an advantage in disabling autocommit mode? From my basic understanding we want to have the autocommit mode enabled at all times and the user can specify the BEGIN command to explicitly start a transaction.

2. Why is the on_error_rollback set to off by default? Wouldn't it make sense to have that on all the time?

Hrm. Even when I set the on_error_rollback set to True it ends up aborting the transaction. I haven't traced the issue yet. Can you please verify if it works for you?

Any updates on this? I've been interested in this feature for quite a while.

@amjith @gma2th any updates on getting this through? Not being able to turn off autocommit for pgcli is basically a dealbreaker for me and many others who want some safety from fat-fingering!