fix: migrate drop the wrong m2m field when model have multi m2m fields

[waketzheng]

Fixes #375
Fixes #150

Description

The two bugs are some thing like the following case:

old = [{'through': 'a'}, {'through': 'b'}, {'through': 'c'}]
new = [{'through': 'a'}, {'through': 'c'}]
print(list(dictdiffer.diff(old, new)))

Got:

[('change', [1, 'through'], ('b', 'c')),
 ('remove', '', [(2, {'through': 'c'})])]

Expected:

[('remove', '', [(1, {'through': 'b'})])]

Reason

When use dictdiffer.diff to compare two lists of dicts, it just compare elements at the same position:

old[0] vs new[0]
old[1] vs new[1]

Solution

• 1. use a custom diff function that generate the expected output:

def diff(old: dict, new: dict, key: str) -> list[tuple]:
      # compare dicts that with same key instead of compare dicts with same index

• 2. reorder the list elements: in this case switch old[1] and old[2] will get the expected output

[henadzit]

Thanks for working on this! I left a few comments.

[henadzit]

Can you please explain why we need to order the fields in this way? Is the code that follows this expects some sort of ordering? What are the criteria for this ordering?

[waketzheng]

1. why we need to order the fields in this way?
   Good question~
   May be a custom diff function is a better solution:

-                if old_m2m_fields and new_m2m_fields:
-                    old_m2m_fields, new_m2m_fields = reorder_m2m_fields(
-                        old_m2m_fields, new_m2m_fields
-                    )
-                for action, option, change in diff(old_m2m_fields, new_m2m_fields):
+                for action, option, change in diff_plus(old_m2m_fields, new_m2m_fields):

2. Is the code that follows this expects some sort of ordering? What are the criteria for this ordering?
   All cases are covered by unittest: https://github.com/tortoise/aerich/pull/376/files#diff-33c13e0b177bacd2f02e29bcb8aea5b49e7ce34901fd8f41fefb65defba1bd33R7

[henadzit]

Thanks for the reply!

I'm not saying that a custom diff function is better, I'm just trying to understand why this ordering is required. if it's really hard to explain with a spoken language, it's a good sign that something isn't right.

[waketzheng]

@henadzit I updated the PR description

[waketzheng]

@henadzit I updated the PR description

@henadzit Thanks for you comments. I created a new PR, which fix the issue by another solution: #390