[PATCH wayland] server: Make serial number 0 mean invalid

[Daniel Stone]

Hi,

On 30 September 2015 at 16:59, Derek Foreman <derekf at osg.samsung.com> wrote:
> Having an invalid serial number is quite useful - for example, we can
> have protocol requests that optionally take a serial number or 0
> instead of having two similar requests.
>
> Signed-off-by: Derek Foreman <derekf at osg.samsung.com>
> ---
>
> Well, let's see where this goes ;)
>
> I've seen a few times now when having an invalid serial number would be
> helpful, is it too late to do this?  Are we going to break anything?
>
> Anything currently doing math on serial numbers can be wrong by one if one
> of the serial numbers has wrapped around - this probably doesn't matter?
>
> I do wonder if we should have a wl_serial_compare() as part of wayland
> but the semantics are tricky (a big serial and a small serial may actually
> have happened quite close to eachother if the small happened after a
> wrap around).
>
> I think the usual comparisons of interest are equality and "is A less than B
> within (arbitrary distance to compensate for wrap-around)".  I think strcmp()
> like semantics won't necessarily be good here because we don't usually need
> something to tell us "b happened before a", more "these almost certainly
> happened in the order specified in the function call".
>
> Where do we go from here?

>From the IRC discussion way back when:
20:27 < krh> if we do need to make a new serial, I'd much prefer to
have it be 1 << 31
20:27 < krh> and then just restrict normal serial numbers space to 0 -
(1<<31 - 1)
20:28 < krh> so that serial number increment is just serial = (serial
+ 1) & 0x7fffffff

I have no relevant opinion on the matter, but I guess that 0x80000000
(or 0xffffffff) is a much more obviously invalid serial number than 0,
which could also just be the victim of something uninitialised.

A helper would be nice, though like you say I think it has to be
directional/multi-return (older inc. wraparound? identical?
newer/error? invalid?). It'd probably be best to do that inside an
active user and only then shuffle it into Wayland.

So, with the change to a different serial, and a note in documentation
somewhere that any extension relying on the invalid-serial behaviour
must itself explicitly document that:
Acked-by: Daniel Stone <daniels at collabora.com>

Cheers,
Daniel