looks like we don't have tests that cover it

Would it be too hard to add a test for that case?

It seems hard, I want to add it, but I haven't thought of a good way yet

I believe we dropped script replication on 7.0 so this is basically a dead code, no?

The code could be alive if we're replicating from an older master, or reading an old (or specially crafted)AOF file

Not sure this is worth the trouble of writing a test or refactoring the code. I'm ok merging it as is.
If we do wanna test it, then generating an AOF with a script is the easiest.

Not sure this is worth the trouble of writing a test or refactoring the code. I'm ok merging it as is.
If we do wanna test it, then generating an AOF with a script is the easiest.

i see in aof.tcl, we have a EVAL timeout with slow verbatim Lua script from AOF test, i can easier reuse it
the another question is, SCRIPT KILL is not a LOADING command, so if we are loading a aof (with busy script), we cannot issue the KILL command. should we change these KILL commands to LOADING? if so, open a new PR to add it?

no, SCRIPT KILL should not be allowed during loading (since the script in that case would be from the AOF file, and we should not kill it).
we should also probably have prevented SCRIPT KILL from killing scripts that were sent from the master, but i don't think we should bother with that now (since we no longer propagate scripts).
i think we can merge this fix just for clean code, under the understanding that it's a very rare issue that in nearly-dead-code.

got it, let's merge it