[PATCH 21/37] dix: Drive tree mark/validate on unmap from paintable not viewable

[Adam Jackson]

On Mon, 2014-10-20 at 19:23 -0400, Peter Harris wrote:
> It's possible I missed a patch or resolved a conflict incorrectly, but
> I'm getting crashes after this patch series.
> 
> Reproducible by running twm and opening/closing the root window popup menu.
> 
> On 2014-10-08 11:04, Adam Jackson wrote:
> > @@ -2742,7 +2742,7 @@ UnmapWindow(WindowPtr pWin, Bool fromConfigure)
> >          return Success;
> >      if (SubStrSend(pWin, pParent))
> >          DeliverUnmapNotify(pWin, fromConfigure);
> > -    if (wasViewable && !fromConfigure) {
> > +    if (wasPaintable && !fromConfigure) {
> >          pWin->valdata = UnmapValData;
> 
> UnmapValData is a special not-pointer that is only valid if
> !pWin->paintable (or !pWin->viewable before this patch series). If
> Composite resets pWin->paintable on this window, miComputeClips will dx
> = pParent->drawable.x - pParent->valdata->before.oldAbsCorner.x; causing
> everything to explode.

I kind of hate UnmapValData for being out-of-band like that.  Seems like
maybe it'd be cleaner to pass along a VTKind?  I also wish mioverlay had
any sort of documentation or internal commentary.

> > @@ -2750,13 +2750,11 @@ UnmapWindow(WindowPtr pWin, Bool fromConfigure)
> >      pWin->mapped = FALSE;
> >      if (wasRealized)
> >          UnrealizeTree(pWin, fromConfigure);
> > -    if (wasViewable) {
> > -        if (!fromConfigure) {
> > -            (*pScreen->ValidateTree) (pLayerWin->parent, pWin, VTUnmap);
> > -            (*pScreen->HandleExposures) (pLayerWin->parent);
> > -            if (pScreen->PostValidateTree)
> > -                (*pScreen->PostValidateTree) (pLayerWin->parent, pWin, VTUnmap);
> > -        }
> > +    if (wasPaintable && !fromConfigure) {
> 
> An extra block here
> 
> +            if (pWin->paintable) {
> +                pWin->valdata = NULL;
> +                pScreen->MarkWindow(pWin);
> +            }
> 
> appears to fix the crash. I haven't considered the equivalent for
> UnmapSubwindows yet.

Hmm yeah okay.  I'll work that in somehow.

- ajax