|author||John Mark Bell <firstname.lastname@example.org>||2008-03-21 00:14:01 +0000|
|committer||John Mark Bell <email@example.com>||2008-03-21 00:14:01 +0000|
Unsightly hack around flickering to white in nsgtk. This needs to be sorted out properly as a matter of urgency.
svn path=/trunk/netsurf/; revision=4005
Diffstat (limited to 'render')
1 files changed, 37 insertions, 0 deletions
diff --git a/render/layout.c b/render/layout.c
index 93df43038..83ea9ca64 100644
@@ -187,7 +187,44 @@ bool layout_block_context(struct box *block, struct content *content)
assert(block->width != UNKNOWN_WIDTH);
assert(block->width != AUTO);
+ /* Why the ifdef? You don't really want to know. If you do, read on.
+ * So, the only way into this function is through the rest of the
+ * layout code. The only external entry points into the layout code
+ * are layout_document and layout_inline_container. The latter is only
+ * ever called when editing text in form textareas, so we can ignore it
+ * for the purposes of this discussion.
+ * layout_document is only ever called from html_reformat, which itself
+ * is only ever called from content_reformat. content_reformat locks
+ * the content structure while reformatting is taking place.
+ * If we call gui_multitask here, then any pending UI events will get
+ * processed. This includes window expose/redraw events. Upon receipt
+ * of these events, the UI code will call content_redraw for the
+ * window's content. content_redraw will return immediately if the
+ * content is currently locked (which it will be if we're still doing
+ * layout).
+ * On RISC OS, this isn't a problem as the UI code's window redraw
+ * handler explicitly checks for locked contents and does nothing
+ * in that case. This effectively means that the window contents
+ * aren't updated, so whatever's already in the window will remain
+ * on-screen. On GTK, however, redraw is not direct-to-screen, but
+ * to a pixmap which is then blitted to screen. If we perform no
+ * redraw, then the pixmap will be flat white. When this is
+ * subsequently blitted, the user gets greeted with an unsightly
+ * flicker to white (and then back to the document when the content
+ * is redrawn when unlocked).
+ * In the long term, this upcall into the GUI event dispatch code needs
+ * to disappear. It needs to remain for the timebeing, however, as
+ * document reflow can be fairly time consuming and we need to remain
+ * responsive to user input.
block->float_children = 0;