Keep deleted messages in cache

[mbaruh]

We want deleted messages to remain in the cache so that the bot can account for them when applying spam rules.

One way of doing that is monkey-patching this part https://github.com/Rapptz/discord.py/blob/master/discord/state.py#L483-L504 but it should be accompanied by proper tests to prevent any funny business on version bumps.

Another way is to cache deleted messages in a separate cache, but well.. it was already cached.

[MarkKoz]

Can you provide more context on this? AntiSpam uses an API call to get message history; it doesn't rely on a cache. Are you also proposing to change it to use a cache?

[mbaruh]

Seems like I misremembered. Then yes, we should use the cache, unless there is some reason against.

The context is that the bot can't apply the duplicates filter to someone spamming invite links, since they are removed.

[MarkKoz]

That's basically a race condition. See #1018

If you only care about solving the race condition, then there's possibly a better way than to cache deleted messages. However, if you want to also account for messages that users delete themselves, then caching them is likely the only option.

[mbaruh]

What would be a better solution? Sebastiaan's suggestion on the other issue doesn't seem to deal with the filters that apply to multiple messages.

[MarkKoz]

Using locks or specifying an order for processing. Anything that involves deleting messages should be done at the end. I don't know if it's actually feasible but it's something to consider.

[mbaruh]

I might not be following. Even if you execute all filters in the right order for a specific message, if it is deleted by one of them, it is possible that in a few seconds it will turn out it's part of a larger stream of spam messages that should be punished. At that point, the message is already deleted and therefore won't be taken into account.

[mbaruh]

I guess you could still call it a race condition, but it's one that is caused by the filters not having information on the user's future behavior, thus the reasoning that we need a cache.

[MarkKoz]

I might not be following. Even if you execute all filters in the right order for a specific message, if it is deleted by one of them, it is possible that in a few seconds it will turn out it's part of a larger stream of spam messages that should be punished. At that point, the message is already deleted and therefore won't be taken into account.

You're right; I was just thinking about it the wrong way.

[Akarys42]

Superseded by #1530