Re: [css-writing-modes] Propose writing-mode: sideways-left

On Sat, Jul 11, 2015 at 12:21 AM, Richard Ishida <ishida@w3.org> wrote:

> [snip]
> i have attached an image that shows various numbered use cases. Fantasai,
> Koji, Florian, could you each indicate how you think CSS properties and
> values should be used to achieve each of the use cases?
>

1, 2, 5: sideways-lr, non-CJKM
3, 6: vertical-rl, CJK
4, 7, 8: sideways-rl, non-CJKM

vertical-lr (Mongolian) is not included in these 8 cases. They look similar
to 3, 6 but lines flow from left to right.

what's the difference between sideways-rl and vertical-rl? Presumably none
> for horizontal latin text.


Almost none. but two I would list are:
1. Ambiguous (unified) characters behave as CJKM for "vertical" and as
non-CJKM for "sideways".
2. There are some subtle typographic differences between Latin characters
in CJK vertical and rotated Latin. They're distinguished between the two
values.


> Or does this rotate all CJKM characters 90º clockwise?
>

Yes for "sideways", no for "vertical". You could consider "sideways" being
no more than rotation, while "vertical" involves CJKM vertical flow
typographic conventions such as their native characters appear in upright.

i'm wondering whether a better name for sideways-lr would be horizontal-bt,
> and possibly horizontal-tb for sideways-rl?
>

I'm not good at bikeshedding, so I'd like to delegate that to someone else
;-)

what i'm not clear about wrt horizontal text that runs up the page is where
> the line start is. Is that defined?


> suppose i wanted to put the following text sideways latin text in a table
> header cell, and that the table cell height is restricted so that it only
> contains three words per line.
>
> "one two three four five"
>
> i'm pretty sure that the words "four five" would be on the second line
> (whether that's to the left or right of the first line), rather than
> appearing on the first and pushing 'one two' to the second line (which
> implies that the latin text has opposite-direction behaviour, and behaving
> like we currently expect rtl text to behave wrt line breaking. Doing that
> with text-orientations sounds odd).
>
> i'm assuming that this is what sideways-lr is for, but if an author used
> that would the word "one" and the word "four" to be touching the bottom of
> the container, or would "three" and "five" be touching the top of the
> container? And where is the line start?
>

I think you're trying to understand in a complex way. Is it easier to
understand if you consider "sideways-rl" is a clock-wise rotation, and
"sideways-lr" is a counter-clock-wise rotation? It's what they do.

if you use Word to add some latin text to a table header that runs up the
> page, it shows the line start at the bottom of the cell. To do that you set
> direction to "Rotate all text 270º".
>
> i can't help thinking that we are struggling because we're trying to do
> two different things with one set of properties.
>

I think opposite. The important distinction is whether you want to rotate
"characters" or "lines". With the proposal, if you want to rotate "lines",
writing-mode is the property, it now allows both 90 and 270. If you want to
rotate "characters" without affecting lines at all, text-orientation is the
property. The current sideways-left and sideways-right are not consistent
in this regard.

i think that where text-orientation comes in is to affect the default
> orientation of vertical text only, ie. CJKM text, either by making the
> latin upright, or by making the CJK text run down the page.  I'm inclined
> to think that support for latin text running up the page or arabic text
> running down it while embedded in CJKM text is something so rare that we
> don't need to worry about it – or at the very least we shouldn't let it
> continue to bog down progress on supporting the basic needs that people
> have been asking for for years now.
>
> i suspect that fitting latin text in a table heading so that it runs
> bottom to top, or top to bottom, is really a case of rotating the
> horizontally laid out text, and nothing more.  (This would mean that only
> 'horizontal-bt' would put a second line on the right, and only
> 'horizontal-tb' would put it on the left.)
>

Yeah, that's where this proposal improves. Now you can use the same
property to rotate clock-wise or counter-clock-wise, just different values.
Non-CJK authors have no question in character orientation once line was
rotated, so all they need to learn is the writing-mode property.


> fwiw, here's an example of Latin text running up the page, however it is
> only in a loose association with the vertical-rl orientation of the chinese
> and japanese alongside it. I can count on one hand the other examples i've
> seen of text running in this direction. I assume that if this were
> translated into Arabic, that would run down the page but with the top of
> the glyphs to the left.
>
> https://www.flickr.com/photos/ishida/19391039009/
>

This example in the current spec is:
  writing-mode: vertical-rl; /* rotate lines (along with characters) 90
degree */
  text-orientation: sideways-left; /* then rotate lines (along with
characters) 180 degree */
With this proposal:
  writing-mode: sideways-lr; /* rotate liens (along with characters) 270
degree */

/koji