[Spice-devel] [PATCH spice-protocol 2/8] Add the StreamMsgGraphicsDeviceInfo message

[Lukáš Hrázký]

On Fri, 2019-01-11 at 05:45 -0500, Frediano Ziglio wrote:
> > On Thu, 2019-01-10 at 14:28 -0600, Jonathon Jongsma wrote:
> > > On Wed, 2019-01-09 at 12:43 +0100, Lukáš Hrázký wrote:
> > > > On Wed, 2019-01-09 at 02:38 -0500, Frediano Ziglio wrote:
> > > > > > 
> > > > > > The message contains information about the graphics device and
> > > > > > monitor
> > > > > > belonging to a particular video stream (which maps to a channel)
> > > > > > from
> > > > > > the streaming agent.
> > > > > > 
> > > > > > Signed-off-by: Lukáš Hrázký <lhrazky at redhat.com>
> > > > > > ---
> > > > > >  spice/stream-device.h | 9 +++++++++
> > > > > >  1 file changed, 9 insertions(+)
> > > > > > 
> > > > > > diff --git a/spice/stream-device.h b/spice/stream-device.h
> > > > > > index 6add42b..77d76af 100644
> > > > > > --- a/spice/stream-device.h
> > > > > > +++ b/spice/stream-device.h
> > > > > > @@ -90,6 +90,8 @@ typedef enum StreamMsgType {
> > > > > >      STREAM_TYPE_CURSOR_SET,
> > > > > >      /* guest cursor position */
> > > > > >      STREAM_TYPE_CURSOR_MOVE,
> > > > > > +    /* the graphics device display information message (device
> > > > > > address and
> > > > > > display id) */
> > > > > > +    STREAM_TYPE_DEVICE_DISPLAY_INFO,
> > > > > >  } StreamMsgType;
> > > > > >  
> > > > > >  typedef enum StreamCapabilities {
> > > > > > @@ -140,6 +142,13 @@ typedef struct StreamMsgData {
> > > > > >      uint8_t data[0];
> > > > > >  } StreamMsgData;
> > > > > >  
> > > > > > +typedef struct StreamMsgDeviceDisplayInfo {
> > > > > 
> > > > > No much documentation, when it is supposed to be sent?
> > > > > In which state?
> > > > > From the agent, from the server or both?
> > > > 
> > > > I'll add some documentation.
> > > > 
> > > > > No capabilities for a new message?
> > > > 
> > > > No, the streaming agent protocol is not stable yet. Better to not add
> > > > a
> > > > capability and make the code more complex if it's not necessary.
> > > > We've
> > > > discussed this before...
> > > > 
> > > > > > +    uint32_t id;
> > > > > 
> > > > > I suppose is not the same "device_display_id", not clear what is
> > > > > it.
> > > > 
> > > > Right, this is basically the equivalent of monitor_id, here it is
> > > > kind
> > > > of the ID of the stream to which the display info belongs. It's
> > > > basically just a sequence of 0, 1, 2, ... and since only a single
> > > > stream/monitor is supported at the moment, only a single message of
> > > > this type is sent and it has id = 0.
> > > > 
> > > > Suggestions how to name it? monitor_id? stream_id?
> > > 
> > > I think stream_id may be ok.
> > 
> > I'll wait for Frediano to confirm he's fine with stream_id as well...
> > 
> 
> Fine with me.
> As a base role I think protocol files should be written thinking about
> and external user. External users should rely on few files (headers if
> specific documentation is not provided). If I look at field name "id"
> just don't say much, just an unique identifier amongst all
> StreamMsgDeviceDisplayInfo. Often people says "look at the code", I
> always find it wrong and an excuse to not document the APIs/protocol.
> If I look at the current code this "id" is the Xrand ID. Is it right?
> No, that's why people should write document and stick to it.

You're right. I knew the name "id" is not descriptive enough when
writing it, but wasn't sure what to use at the time and never got back
to fix it.

> > > > > > +    uint32_t device_display_id;
> > > > > > +    uint32_t device_address_len;
> > > > > 
> > > > > No limit? Is 4gb fine? I suppose you want a length to be able to
> > > > > extend the message in the future, otherwise you can use the size
> > > > > of the message.
> > > > 
> > > > How do you want me to set a limit here? In a comment?
> > > > 
> > > > I initially had these wrapped in a message that contained an array of
> > > > these, hence the length. It is still unclear how this will exactly
> > > > work
> > > > with multi-monitor support. This message is likely going to change
> > > > with
> > > > that... Should I change it now?
> > > 
> > > Personally I would not try to accomodate multimonitor until we know
> > > what we need.
> > > 
> > > > 
> > > > > > +    uint8_t device_address[0];  // a zero-terminated string
> > > > > 
> > > > > Encoding? Utf-8?
> > > > 
> > > > The contents are the address of the device, I don't think utf really
> > > > matters here.
> > > 
> > > Well, the address format is defined as
> > > 
> > > "pci/<DOMAIN>/<SLOT>.<FUNCTION>/.../<SLOT>.<FUNCTION>"
> > > 
> > > So, aside from the string "pci/", the rest is just hex numbers and
> > > dots. So we don't need to encode a bunch of special characters, just
> > > characters within the ASCII range. But I don't suppose it would hurt to
> > > document it as utf-8...
> > 
> > Yes. I would aslo assume utf-8 is the default and there hardly is any
> > other consideration anyway. If it's not explicitly stated anywhere, I
> > would just document strings in general being utf-8 encoded in the
> > protocol and not document it per string?
> > 
> 
> From the header file "donald duck\0" is a perfect device address.
> Maybe a comment about the format (even a reference to something
> else but I would be better if people don't have to go crazy to
> get to it)

I've realized that as well. It's described on the spice server API
functions for QXL interface, but this is a different place, I think
I'll duplicate the information here.

Cheers,
Lukas

> > Cheers,
> > Lukas
> > 
> 
> Frediano
> 
> > > > > I suppose by "zero-terminated" you mean if not terminated is a
> > > > > protocol violation.
> > > > 
> > > > Yes. Do you prefer something else?
> > > > 
> > > > Cheers,
> > > > Lukas
> > > > 
> > > > > > +} StreamMsgDeviceDisplayInfo;
> > > > > > +
> > > > > >  /* Tell to stop current stream and possibly start a new one.
> > > > > >   * This message is sent by the host to the guest.
> > > > > >   * Allows to communicate the codecs supported by the clients.
> > > > > 
> > > > > Frediano