being clearer about css-wide keywords and reset-only sub-properties [css-cascade][css-grid][css-border]

https://drafts.csswg.org/css-cascade/#shorthand says:
  In other cases, a property might be a reset-only sub-property
  of the shorthand: Like other sub-properties, it is reset to its
  initial value by the shorthand when unspecified, but the shorthand
  might not include syntax to set the sub-property to any of its
  other values. For example, the border shorthand resets
  border-image to its initial value of none, but has no syntax to
  set it to anything else. [CSS3BG]

  If a shorthand is specified as one of the CSS-wide keywords
  [CSS3VAL], it sets all of its sub-properties to that keyword,
  including any that are reset-only sub-properties. (Note that these
  keywords cannot be combined with other values in a single
  declaration, not even in a shorthand.)

This principle -- that the CSS-wide keywords always affect all
subproperties of a shorthand -- has, I believe, been well-known to
members of the working group for a long time.  However, it hasn't
been clear to some implementors based on the prose elsewhere that
actually defines these properties.

I was led to this issue based on the discussion starting in
https://bugzilla.mozilla.org/show_bug.cgi?id=1176792#c33 , which is
about implementing the grid-{column,row}-gap properties, which
https://drafts.csswg.org/css-grid/#grid-shorthand defines as being
reset by the grid shorthand but not otherwise settable by them:

  The grid property is a shorthand that sets all of the explicit
  grid properties (grid-template-rows, grid-template-columns, and
  grid-template-areas), all the implicit grid properties
  (grid-auto-rows, grid-auto-columns, and grid-auto-flow), and the
  gutter properties (grid-column-gap and grid-row-gap) in a single
  declaration. If <‘grid-auto-rows’> value is omitted, it is set to
  the value specified for grid-auto-columns. Other omitted values
  are set to their initial values.

Mats though that this meant 'grid: inherit' would need to set these
properties to their initial values, rather than to 'inherit', partly
because he wasn't aware of the text from css-cascade that I quoted
above.

When I pointed out that this concept was present elsewhere, Mats
pointed out a testcase that shows that Chrome gets this wrong for
'border' resetting 'border-image', and appears to do what Mats
thought the spec wanted him to do for 'grid':
  https://bug1176792.bmoattachments.org/attachment.cgi?id=8688433
I believe the Gecko rendering here is correct (multiple sets of
green corners, since border-image is inherited) and the Chromium
rendering is not (one set of green corners, and a black border
inside).


I think this points to lack of clarity in the spec wording, given
that the definition I quote above in css-cascade is not findable
from the definitions of these reset-only sub-properties.  I think
that the prose defining these reset-only sub-properties in css-grid,
css-border, etc., should:

  1. link to the definition of the term "reset-only sub-property" in
     css-cascade

  2. either (a) be more deferential to that definition about what it
     means (and not state that the properties are reset to initial
     values) or (b) explicitly mention that CSS-wide keywords are
     applied rather than resetting those subproperties to their
     initial values (as all other values of the shorthand do).  Or,
     to put it another way, spec prose should not state that most
     values reset the subproperty to its initial value without also
     stating that the css-wide keywords do otherwise.

-David

-- 
𝄞   L. David Baron                         http://dbaron.org/   𝄂
𝄢   Mozilla                          https://www.mozilla.org/   𝄂
             Before I built a wall I'd ask to know
             What I was walling in or walling out,
             And to whom I was like to give offense.
               - Robert Frost, Mending Wall (1914)