On Tue, Jan 30, 2024 at 03:30:13AM +0000, JeeHeng Sia wrote: ... > > Sharing code is good, but if we have to parametrize the entire table, then > > we might as well keep Arm and RISCV separate. Building the table first > > with this struct, just to have it built again with the build_append API, > > doesn't make much sense to me. Do Arm and riscv really diverge on all > > these parameters? If not, then just add the parameters which do diverge > > build_scpr's arguments. > It is kind of chicken and egg thing, I would suggest let the arch code to > fill in the value. It doesn't make sense to change again in the future when > both riscv and arm realized the parameters were different. > Can arm confirm that these values wouldn't change in the future? > >
We can't be sure that arm nor riscv will change in the future, but we (arm/riscv/etc. QEMU developers) control the code for both, so I don't see a problem with only parametrizing what's necessary today and then extending that, or completely separating, later if necessary. In any case, I'd rather see the two completely separate from the start, than to see the structure with all the parameters get added. There's no difference to me when reading a list of 's->param_name = value' or a list of build_append_int(table, value, size) /* param_name */. And, given we need the later eventually anyway, then there's no reason for the former at all. Thanks, drew
