[css-content] Clarify valid and default values of `string-set`

[Crissov]

In the 2003 WD of CSS Content, it was anticipated that the string-set property would take as its value a list of pairs of custom ident and source:

The 'string-set' property accepts values similar to the 'content' property, including the extraction of the current value of counters.

META[author] { string-set: author attr(author); }
HEAD > TITLE { string-set: title contents; }

All subsequently released drafts of that module, including the current ED, limited the capability of this property in prose. It is supposed to effectively always store the equivalent of content() or content(text):

The string-set property copies the text content of an element into a named string, …

However, its syntax requires at least one <string> after the initial custom identifier in order to select what should be stored in the named string:

Value: none | [ <custom-ident> <string>+ ]#

Accordingly, some of the following examples contain functions that return a <string>:

h2 { string-set: heading content() }

h1 { string-set: header content(before) ':' content(text); }

section { string-set: header attr(title) }

However, the very first example contains the keyword contents in place of a <string>:

H1 { string-set: chapter contents; }

This keyword is specified in section 2.3 of the same module as part of the <content-list> type, which is strongly tailored to the content property.

<content-list> = [ <string> | <image> | <attr()> | contents | <quote> | <leader()> | <target> | <string()> | <content()> | <counter> ]+

In conclusion, please clarify the default and valid values of the string-set property, which might look like one of these:

1. none | <custom-ident> – suggested by current prose
2. none | <custom-ident># – also consistent with current prose
3. none | [ <custom-ident> <string>* ]# – compromise between current syntax and prose
4. none | [ <custom-ident> <string>+ ]# – current syntax
5. none | [ <custom-ident> <content-list>+ ]# – suggested by examples
6. none | [ <custom-ident> [ <string> | <counter> | <attr()> | <content()>]* ]# – covers all examples except the first which would be changed to use content() instead, also includes all types available in the fallback "/ alt" value of the content property

[liZe]

Hi @Crissov,

Good to see someone else interested in this topic!

In WeasyPrint we do implement 5. We’re pretty happy with this choice, that gives a large amount of possibilities and uses the already defined <content-list>.

(I think that the "+" after <content-list> is useless, isn’t it?)

[Crissov]

Indeed, <content-list> already ends with +, so it is redundant in my option 5.

However, I’m not sure whether string-set should really also support <image>, which is part of <content-list>. Therefore I did not include it in option 6. I’m less sure about omitted <quote> and <leader()>. Adding expensive <target> and (possibly circular) <string()> there seems also unnecessary to me.

[liZe]

I agree, 6 is probably enough for users and will avoid useless problems for implementations.

(I never actually tried <image>… and it just doesn’t display it in WeasyPrint.)