[Spice-devel] [PATCH 00/17] WIP: Refactor the streaming agent towards a more standard C++ style

Tue Feb 20 14:46:49 UTC 2018

> On 20 Feb 2018, at 15:36, Lukáš Hrázký <lhrazky at redhat.com> wrote:
> 
> On Tue, 2018-02-20 at 15:22 +0100, Christophe de Dinechin wrote:
>>> On 19 Feb 2018, at 18:29, Lukáš Hrázký <lhrazky at redhat.com> wrote:
>>> 
>>> On Fri, 2018-02-16 at 17:15 +0100, Christophe de Dinechin wrote:
>>>> From: Christophe de Dinechin <dinechin at redhat.com>
>>>> 
>>>> The streaming agent started as C code. This series converts
>>>> the style to something that is more usual for C++, notably:
>>>> 
>>>> - Adds encapsulation and RAII for resources such as the stream
>>>> - Splits functionality into several classes with clear roles
>>>> - Puts message formatting and writing in a reused based class
>>>> - Isolate what's specific to each message in three derived classes
>>>> - Isolate X11-specific code in separate classes, one for cursor messages,
>>>> one for the thread polling the data.
>>>> 
>>>> Reasons for marking this WIP:
>>>> 
>>>> 1. This series is, unfortunately, not correctly tested, because on my
>>>>  machine, I currently have a black screen with the 'master' streaming
>>>>  agent, server, protocol, common, plugin and spicy. Fallback to MJPEG
>>>>  leads to a large repetition of messages like:
>>>> 
>>>>  (spicy:4217): GSpice-CRITICAL **: need more input data
>>>> 
>>>>  What I have tested using the -d option is that the syslog output
>>>>  from the agent looks similar (size of data captured, etc) relative
>>>>  to master both for the MJPEG plugin and with fallback. But without
>>>>  a picture, I am still concerned about the lack of testing.
>>> 
>>> Christophe already knows, but in case anyone will benefit from this,
>>> it's the incomplete frame bug, fix should be on the ML here:
>>> 
>>> [Spice-devel] [PATCH spice-server 8/8] stream-channel: Send the full
>>> frame in a single message
>>> 
>>> Christophe, I've tested your patches, it's streaming, but the cursor
>>> disappears after a while, so something's not entirely right.
>> 
>> So there is probably an issue with the cursor message… I will look into it.
> 
> See my reply to 12/17 of the series, should probably be the issue.
> 
>> What guest type (e.g. f25, rhel…)? I’ve not seen that. I’ll try again and see if I can reproduce.
> 
> f27, should not be related.
> 
>>> 
>>>> 2. The classes were isolated, but not moved in separate headers. This
>>>>  is intentional. I prefer to make sure that the changes on the code
>>>>  are agreed on before we move the individual classes to their own
>>>>  headers. It thinks it will also faclitate history browsing
>>> 
>>> I'd like to point out the classes are not entirely isolated, there is
>>> still the static bool streaming_requested (besides the necessary
>>> quit_requested, although I have an idea for that), which ties together
>>> the ConcreteAgent::CaptureLoop and Stream::read_command_from_device.
>> 
>> Interesting. That will force me to re-review the whole thing, because I definitely remember moving streaming_requested within the Stream class, but I don’t see that anymore. Unless I ran into some issue and decided to postpone that?
> 
> Must've gotten lost... But the solution to put client_codecs and
> supposedly the streaming_requested is honestly ugly :)

Well, I really only put outfgoing messages in their own classes. Incoming messages remain to be done. That being said, the logic is really small, so I though that was OK for the moment.

> and seems like you started taking shortcuts :)

No, simply doing things in rough order of priorities.

> I was actually working on that part myself, so it turned out ok as I have some rough solution, which I’ll need to incorporate with your design, if I can manage that :)

Cool. So I won’t touch that part then.

> 
>>> 
>>> I have an unfinished patch for this, which will need heavy rebasing. In
>>> case you are/will be looking into it, Christophe, let me know.
>> 
>> My reading is that you have a patch for the coupled case (quit). Is it OK if I add one for streaming_requested to the next spin, or do you plan to do that?
> 
> I don't have the patch yet, just an idea for the quit. As for
> streaming_requested, see above. I'd like to do that, on top of your
> patches, once the dust settles a bit...
> 
>>> 
>>> Lukas
>>> 
>>>> 3. This series compiles without warnings both with GCC in C++11 mode
>>>>  and by clang in gnu++11 mode. However, Frediano pointed out that it
>>>>  uses a designated intiializer syntax that is presently a GNU
>>>>  extension (the C99 designated initializer syntax). We may consider
>>>>  it a problem or not. If it's a problem, it's sufficient to remove
>>>>  the designators, but I think they add to readability.
>>>> 
>>>> 4. Our current configure.ac requires a warning if all initializers are
>>>>  not present. Since padding was made explicit in the protocol, this
>>>>  requires the code to initialize padding fields, which I don't like.
>>>> 
>>>> 5. The series integrates off-topic patches sent in a separate series, but
>>>>  which are necessary to successfully build with both clang and gcc.
>>>> 
>>>> This series can also be browsed at
>>>> https://gitlab.com/c3d/spice-streaming-agent/merge_requests/1/commits
>>>> 
>>>> Christophe de Dinechin (17):
>>>> Add missing <string> header
>>>> log_binary is really a boolean
>>>> Eliminate signed/unsigned warning
>>>> Do not create std::string for constants
>>>> Use RAII to cleanup stream in case of exception or return
>>>> Replace inefficient C-style initialization with C++-style
>>>> Get rid of C-style memset initializations, use C++ style aggregates
>>>> Use C++ style for cursor message initialization instead of C style
>>>> Reorder headers according to style guide
>>>> Since we use a namespace, simplify name of local classes
>>>> Move read, write and locking into the 'Stream' class
>>>> Convert message writing from C style to C++ style
>>>> Add more meaningful syslog reporting
>>>> Create a class encapsulating the X11 display cursor capture
>>>> Create FrameLog class to abstract logging of frames
>>>> Remove client_codecs global variable, moved inside the 'Stream' class
>>>> Move the capture loop in the ConcreteAgent, get rid of global agent
>>>>   variable
>>>> 
>>>> src/concrete-agent.cpp        |   1 +
>>>> src/concrete-agent.hpp        |   4 +
>>>> src/mjpeg-fallback.cpp        |   2 +-
>>>> src/spice-streaming-agent.cpp | 521 +++++++++++++++++++++++++-----------------
>>>> 4 files changed, 311 insertions(+), 217 deletions(-)
>> 
>> 
> _______________________________________________
> Spice-devel mailing list
> Spice-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/spice-devel