[vlc-devel] [vlc-commits] Implement a first try for the WindowLess on Linux

[Rémi Denis-Courmont]

This patch cannot be left like this without follow-up. It's a massive
performance regression, and introduces or reintroduces bugs.

On Thu, 13 Dec 2012 15:19:23 +0100 (CET), git at videolan.org (Cheng Sun)
wrote:
> +bool VlcWindowlessLinux::handle_event(void *event)
> +{
> +#warning FIXME: this is waaayyy too slow!
> +    XEvent *xevent = static_cast<XEvent *>(event);
> +    switch (xevent->type) {
> +    case GraphicsExpose:
> +        XGraphicsExposeEvent *xgeevent =
> reinterpret_cast<XGraphicsExposeEvent *>(xevent);
> +        Display *display = xgeevent->display;
> +
> +        int screen = XDefaultScreen(display);

This is wrong. The screen should correspond to the drawable, as specified
in XPutImage() below.

> +        XVisualInfo visual;
> +        XMatchVisualInfo(display, screen, 24, TrueColor, &visual);

This function call can fail.

> +        XImage *image = XCreateImage(display, visual.visual, 24,
ZPixmap,
> +                                    0, &m_frame_buf[0],
> +                                    m_media_width, m_media_height,
> +                                    DEF_PIXEL_BYTES*8,
> +                                    m_media_width * DEF_PIXEL_BYTES);
> +
> +        const NPRect &clip = npwindow.clipRect;
> +        XPutImage(display, xgeevent->drawable,
> +                  XDefaultGCOfScreen(XScreenOfDisplay(display,
screen)), image,
> +                  clip.left - npwindow.x, clip.top - npwindow.y,
> +                  clip.left, clip.top,
> +                  clip.right - clip.left, clip.bottom - clip.top);

XShm functions would be vastly more efficient when rendering locally.


And then after Laurent and I have spent so much efforts moving away from
Xlib, I cannot accept reverting without justifications. Unlike with VAAPI,
GLX, VDPAU or Qt4, I fail to see any reason why XCB cannot be used here.

-- 
Rémi Denis-Courmont
Sent from my collocated server