Actually, the reason it has to do with legacy video drivers appears to be a limit in the size or complexity of a region.
In previous Mozilla versions, invalidation was done using rectangles, but that view manager change made invalidation use platform regions. There appears to be a limit to either the size or complexity of regions on legacy video drivers. Mike Kaply Mikus Grinbergs wrote: > On Tue, 11 Dec 2001 09:36:20 -0600 Michael Kaply <[EMAIL PROTECTED]> >wrote: > > This problem came back when some changes were made to the view manager > > (cross-platform) > > > > At this point we believe it is a limitation in legacy video drivers. We > > are continuing to investigate. > > I hear what you are saying, but I have DIFFICULTY understanding > how the video driver might be responsible. The overlay problem > occurs with long pages, but NOT with short pages. Yet the > __same__ video driver is being called in both instances. > > To me, it seems far-fetched that the __driver__ would care how > many more characters there are in the page than are shown on the > screen. It is __Mozilla__ that has to "organize" the data for > the ENTIRE page (whether long or short). Therefore I suspect > that Mozilla is doing something DIFFERENT when its page is long. > > For me, the overlay problem exhibits "banding" -- meaning that > in a very long page, there are offsets-from-the-beginning where > overlay *does* occur, and offsets-from-the-begginning where it > does *not* occur. [What the problem seems to be is that previous > text-color pixels are *not* "erased" from the screen before the > pixels from the new text position are "added" to the screen.] > > Is Mozilla properly passing data addresses (segment+displacement), > for certain combinations of which the driver misbehaves? Maybe. > [But note that in late 0.9.5 the overlay problem *was* fixed, > without having to change the driver.] > > Is Mozilla passing unexpected values (eg, segment+displacement), > or failing to pass certain directives (eg, "redraw background") > for certain offsets when the web page is long? Also, maybe. > > mikus
