<html> <head> <base href="https://bugs.freedesktop.org/" /> </head> <body> <div> <a class="bz_bug_link bz_status_NEW " title="NEW --- - GIMP image rendering is broken with --enable-xlib-xcb" href="https://bugs.freedesktop.org/show_bug.cgi?id=67505#c3">Comment # 3</a> on <a class="bz_bug_link bz_status_NEW " title="NEW --- - GIMP image rendering is broken with --enable-xlib-xcb" href="https://bugs.freedesktop.org/show_bug.cgi?id=67505">bug 67505</a> from <a class="email" href="mailto:psychon@znc.in" title="Uli Schlachter <psychon@znc.in>"> Uli Schlachter</a> <pre>Easily reproducable with gimp 96a090d9b12d20deec20248a841f7b36e103c17c. No idea what to do about this. I say the bug is in gimp, Chris blames cairo-xcb. The problem goes something like this: - gimp draws to a small part of an image surface - gimp wants this part to be visible in the GUI and uses it as source surface in cairo_paint() - cairo-xcb uploads the whole image surface to the X11 server (which is way more than is actually needed) - Since this was an expensive operation, cairo-xcb attaches a so called snapshot to the image surface. This snapshot keeps alive the temporary pixmap that the image surface was uploaded to. <time passes> - gimp wants to redraw another part of the visible image and draws to another part of the above image surface - cairo-xcb sees the snapshot attached to the image surface and optimizes away the new upload, because it still has the pixmap that the surface was uploaded to before. Thus, old data is used. I blame gimp because IMHO the API docs imply that it should call cairo_surface_flush() before directly modifying an image surface and mark_dirty() afterwards. The flush() would get rid of the snapshot. Gimp currently only does these calls the first time that it directly modifies the surface. Chris argues that gimp never touches the same area of the image surface twice and that it calls flush();mark_dirty() when all of the temporary image was used. So the bug is that cairo-xcb uploads more of the image surface than it really has to. Also, it needlessly uploads to a temporary pixmap when it could paint directly to the target surface. (And such an optimization would fix the visible results of this bug and thus hide it) If I add a call to cairo_surface_mark_dirty(xfer); to gimp_display_shell_render() (after xfer was modified, gimp dies: gimp-2.9: cairo-surface.c:1585: cairo_surface_mark_dirty_rectangle: Assertion `! _cairo_surface_has_snapshots (surface)' failed (This basically says: You should have called cairo_surface_flush(), so that the snapshot is detached) If I remove the call to _cairo_surface_attach_snapshot() from _cairo_xcb_surface_picture(), the problem (obviously) goes away (and cairo-xcb re-uploads the whole image surface on each cairo_paint()).</pre> </div> <hr> You are receiving this mail because: <ul> <li>You are the QA Contact for the bug.</li> </ul> </body> </html>