__x86_64__ is defined on the on the x32 architecture and the conditionals in
debug.c therefore assume the size of (void*) etc:
debug.c: In function 'getMcontextEip':
debug.c:757:12: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
return (void*) uc->uc_mcontext.gregs[16]; /* Linux 64 */
^
debug.c: In function 'logRegisters':
debug.c:920:21: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
logStackContent((void**)uc->uc_mcontext.gregs[15]);
We can remedy this by checking for __ILP32__ first. See:
https://wiki.debian.org/ArchitectureSpecificsMemo
... for more info.
Soon or later we may have code in freeClient() that may have to deal
with ACLs. Imagine for instance the command proposed multiple times (not
sure if this will ever be accepted but still...):
ONCLOSE DEL mykey
Accumulating commands to run when a client is disconnected. Now the
function is compatible with such use cases.
Related to #5829.
We can't trust modules commands flagging, so module commands must be
always explicitly added, with the exception of +@all that will include
everything. However something like +@readonly should not include command
from modules that may be potentially dangerous: our categories must be
safe and reliable and modules may not be like that.
It does not make much sense to limit what modules can do: the admin
should instead limit what module commnads an user may call. So
RedisModule_Call() and other module operations should be able to execute
everything they want: the limitation should be posed by the API exported
by the module itself.
