The wall of changing the Django service from Python 2.7 to Python 3

Chira back class. It's sad to forget

Premise

A long time ago, something that I wrote properly in Python 2.7 is still alive, so I fixed it. It's a pain because I'm doing bad practices that I don't do now.

Celery

Deprecated when djcelery notices?

From around Celery 3.1 (now Celery 4.0 series?) Celery itself supports Django.

It's good in itself, but the method has changed a little, and it's a sensuous expression, but the flow of "reading the Celery settings after loading the Django environment" is "to read the Django module after starting Celery". Maybe it's because of the "go" order, I end up putting modules in the execution directory

• / opt / proj ... Django's project storage. Here is proj /
• / home / proj ... worker's directory

At some point, I feel like I'm confused if there is no proj / under / home / proj. Or add the module load path after starting Celery?

social-auth

Instructions to use social-auth-app-django.

I had the impression that the migration path was relatively well written, but it must have been a fairly destructive change, and all module names were changed.

https://github.com/omab/python-social-auth/blob/master/MIGRATING_TO_SOCIAL.md

Relative import

In a certain place, the local file (module) name is specified as a character string in the flow of performing the processing equivalent to __import__ (). This was not supplemented in 2to3. Since it does not contain ., it means "not", and as a result, it is assumed that it is not fully loaded. Another process ran, which caused a lot of trouble.

String, bytes

In the first place, it may be possible to write without clearly distinguishing str and unicode Python 2, Suddenly there is a part of confusion when asked to make it strict.

• I'm ashamed of myself because the data wasn't handled properly
• It's confusing that the external modules (including those of CPython) are sloppy

There are many cases where Popen () is used on the system, but it is revealed that it is sloppy due to the character encoding such as stdout. Especially when throwing from stdout to yet another function, is it bytes or str?

I wonder if there was any confusion this time for those who have strict external modules. In the past, I remember eating the csv module and other basic things. I have no choice but to adapt to the new method after being confused.

Python 3.4 -> Python 3.5

The experimental environment is Ubuntu 14.04 and the other environment is Ubuntu 16.04, and I was angry with ʻopen (filename,'b') when migrating. ```open (filename,' rb') ``.

Summary

This time, the impression was that the disaster that hadn't been maintained for a while was bigger than Python 2-> 3.