Paolo Bonzini <pbonz...@redhat.com> writes:
> On 03/11/20 17:25, Daniel P. Berrangé wrote:
>>> OptsVisitor, StringInputVisitor and the keyval visitor have
>>> three different ideas of how a human could write the value of
>>> a boolean option. Pay homage to the backwards-compatibility
>>> gods and make the new common helper accept all four sets (on/off,
>>> true/false, y/n and yes/no), and case-insensitive at that.
>>>
>>> Since OptsVisitor is supposed to match qemu-options, adjust
>>> it as well.
>> FWIW, libvirt does not appear to use true/false or y/n, nor
>> ever use uppercase / mixed case.
>>
>> IOW this level of back compat may well be overkill.
>>
>> I'd particular suggest deprecating case-insensitivity, as
>> Yes, YES, yEs feel unlikely to be important or widely used.
>
> True; at least it's type-safe code unlike the short-form boolean option.
> It only hurts in the odd case of a boolean option becoming on/off/auto
> or on/off/split.
Another argument for deprecating values other than "on" and "off".
> I didn't want to introduce deprecation at this point, because
Not wanting to rush deprecation this close to the release is a valid
point. So is not wanting to rush in additional sugar :)
> consistency is better anyway even if we plan to later deprecate
> something. For example, since there is a common parser now, introducing
> deprecation would be much easier. It also lets us switch parsers even
> during the deprecation period (which is how I got into this mess).
On the other hand, we'll have more to deprecate. Status quo:
on/off yes/no y/n true/false case-sensitive?
qemu_opt_get_bool() X no
keyval visitor X no
string visitor X X X yes
opts visitor X X X no
For once, there is no way to blame QemuOpts ;)
qemu_opt_get_bool() is everywhere in CLI and HMP.
The keyval visitor is used for QAPIfied human-friendly interfaces: block
stuff, -display, -audio.
The string input visitor is used for -object, -global, HMP object_add,
qom-set, migrate_set_parameter.
The opts visitor is used for -mon and its sugared forms (new in 5.2!),
-net, -netdev, -acpitable, -numa.
I'd very much prefer to deprecate the odd ones in string and opts
visitor, and not copy them elsewhere.
We can still factor out a common parsing function if we like: pass a
flag that makes it recognize deprecated forms. A single flag should
suffice; it lets odd ones spread from string to opts visitor and vice
versa, which I find tolerable.
