- From: Tab Atkins Jr. <jackalmage@gmail.com>
- Date: Wed, 26 Jan 2011 19:28:18 -0800
- To: "Eric A. Meyer" <eric@meyerweb.com>
- Cc: www-style@w3.org
On Wed, Jan 26, 2011 at 6:45 PM, Eric A. Meyer <eric@meyerweb.com> wrote: > At 4:52 PM -0800 1/26/11, Brad Kemper wrote: > >> On Jan 26, 2011, at 1:32 PM, fantasai <fantasai.lists@inkedblade.net> >> wrote: >> >>> >>> It was undefined what happened if you overlapped, so I put in a >>> definition that >>> seemed to make sense? Overlapping sides means negative middles, so I >>> just floored >>> the middles at zero. >> >> Hmm. Is it too late too late to make a change like that, to allow overlap >> and still interpret negative middles as zero? It does seem like a fairly >> easy win, if it was implemented that way. You'd have all your corners and >> sides... > > Okay, now I'm seeing the problem. (So I didn't fully understand > 'border-image' after all!) The problem is that you really are slicing the > image up into nine adjacent regions, not copying portions of the image into > eight border slices and a background. For example, the left-side slice has > to be the region bounded by the bottom of the top left corner slice, the > left-offset value, and the top of the bottom left corner slice. Yes. (Though it could be changed, I like the physical connection that this metaphor has.) > (Aside: even now, how are UAs supposed to treat '50%' when the image has > an odd-number-of-pixels dimension? What slices result?) Rounding is always UA-defined. Don't do that if you want a predictable result. > The only way I can see to rescue what I was proposing is to say that where > slice areas overlap, the overlapping region is what's used for some slices. > Specifically the top, right, bottom, left, and fill slices. So given this > image: > > 123 > 234 > 345 > > ...then 'border-image-slice: 2;' would yield the following slices > > 12 2 23 > 23 3 34 > > 23 3 34 > > 23 3 34 > 34 4 45 > > That's not entirely intuitive either, though. It would yield the result of > '100%' (or, in this case, '3') causing every slice to use the full image, > but the cases in the range 51% - 99%, like this one, are a little weird. > And they could lead to some odd joins between border regions. > Unless that sort of result is seen as acceptable or even desirable, then I > think Tab's right and the only way to make the "repeat a single symbol > around the edge" use case work is with a new keyword. Either that or it's > back to 3x3 grids for everything and a number of years repeatedly explaining > why something so apparently simple needs a kinda goofy-looking base image to > drive it. Agreed; the 100% case isn't too bad, but the other cases are confusing and weird. We can add this in B&B 4 at the same time we do 5x5 grids, like Brad suggests. ~TJ
Received on Thursday, 27 January 2011 03:29:11 UTC