I think the more severe is when we I have a user like user hello &abc.
The server will refuse to start.

@huangzhw but why would such a thing happen?

• The new ACL SAVE will never generate such a thing.
• The old ACL SAVE would, but that would have lead to a crash in earlier versions, so i can conclude users aren't facing this scenario.
• Someone may write that line manually into his config or ACL file, and that would work with one config of acl-pubsub-default, and not with the other. but if that's a manual edit, it's the same as many other invalid configuration combinations.

bottom line:

• i don't mind failing an ACL command, an interactive user session will realize what he did wrong and issue it again.
• i don't mind failing to start redis right after a manual edit to the acl or config file.
• i do mind not being able to restart after ACL SAVE or CONFIG REWRITE
• i do mind not supporting old config files that used to work in the past.

i think this simple one line change fits all of these..

If you have a old server, execute config rewrite and stop server, then replace binary, start will fail.

If you have a old server, execute config rewrite and stop server, then replace binary, start will fail.

right. but if you do the same without replacing the binary, start will fail too.
that CONFIG REWRITE was broken, and i think it's safe to assume no one used it this way (if they did they would have already noticed and fixed it).

point is, we're not breaking compatibility with (a valid) old version of the config.

Agree. You are right. I think this fix is enough.

folded into #8794