Unnecessary changelog records for tag add/change/remove

[netsandbox]

Environment

• Python version: 3.6.9
• NetBox version: 2.10.3

Steps to Reproduce

1. create two tags TagA and TagB
2. create a virtual-machine (or device)
3. add TagA to the virtual-machine
4. remove TagA and add TagB to the virtual-machine (in one request)
5. remove TagB from the virtual-machine

Expected Behavior

Steps 3., 4. and 5. create only one changelog record

Observed Behavior

Step 3. creates two changelog records (with the same request id), where only one record shows the tag add and one shows no change.

Step 4. creates three changelog records (with the same request id), where one record shows the TagA remove, one the TagB add, and one shows no change.

Step 5. creates two changelog records (with the same request id), where only one record shows the tag remove and one shows no change.

Here are the database extras_objectchange records for the above steps 2.-5.:
https://gist.github.com/cloos/e7dd4c3b1f064098e4250e7cddb29c85

[jeremystretch]

Multiple change records are created because tags (a many-to-many relationship) are added and removed using separate queries. I agree that it's not optimal but the necessary data is still captured. Attempting to consolidate these changes would likely take quite a bit of effort. If you'd like to volunteer for this work, I'll assign it to you. Otherwise, I don't think there's much to be done.

[netsandbox]

I think it is ok for Step 4. to have two change records, one for tag remove and one for tag add.

But the additional change records in Step 3., 4., and 5. which shows no changes should not be created.
It's a little bit annoying when you try to inspect the object changes and always have change records which shows no object changes. Also if you frequently change tags, this unnecessarily increases the extras_objectchange table.

Sadly my knowledge of the NetBox code is not good enough to fix this myself in a reasonable time.

[patrickfnielsen]

I agree with @netsandbox, there's no reason to create a log entry if there's no changes.
The best solution I could find with my limited knowledge of the system, is to check if there's any diff between the objects before creating a ObjectChanges - This has the drawback of requiring an extra db call.

@jeremystretch Is this something that you would accept a PR for? In my limited testing it seems to fix the problem, but I'm not sure if theres any other drawbacks of doing it this way.

[jeremystretch]

With the improvements made to change logging in v2.11 (see #5913), we should be able to update the first ObjectChange instance on m2m_changed instead of creating a second instance. I'm going to dig into this a bit more.