ODD processing for glosses and descs

[martindholmes]

Arising from ATOP discussions with @sydb and @HelenaSabel:

If gloss[parent::attDef[@mode="change"]], what should happen? Should this gloss be deemed to replace some existing gloss -- and if so, if there are multiple <gloss>es, then which one? The one with the same @xml:lang? And if there isn't one with the same @xml:lang, should the new <gloss> just be added?

And in the case of <remarks>, should we be matching the combination of @xml:lang and @ident? (This leaves aside for the moment the question of <exemplum>, which we believe should also have @ident and @mode to potentially enable the same mechanism, but which has other unsolved problems being handled in #1933.)

[peterstadler]

From my understanding (and the practice in the TEI Guidelines) there's always just one desc/gloss/remarks for each language (with the exception of <desc type="deprecationInfo"> which is to be excluded from this rule).
My proposal would be to make this assumption explicit in the Guidelines and proceed with the ODD processing as proposed by @martindholmes : a @mode="change" on the parent element would result in a replacement of the respective language if present, else it would be added.

[sydb]

I think @peterstadler’s idea is the right approach, at least for P5. Have not thought through if there are wider implications, though.
In the meantime, I checked: Of the 851 elements in p5.xml that have a <gloss> child, none has 2+ with the same value of @xml:lang. (I did not check if any have 2+ that do not have @xml:lang, because I don’t think we care about those.)

I am assigning to myself to write a constraint (in p5odds.odd unless Council says otherwise) to enforce this.
I am hoping someone else self-assigns to make sure the Stylesheets do the right thing.

Note-to-self
In pure XPath: for $gp in //t:*[t:gloss] return concat( local-name($gp), ' of or in ', $gp/ancestor-or-self::*[@ident][1]/@ident, ': ', count( $gp/t:gloss/@xml:lang ), ', ', count( distinct-values( $gp/t:gloss/@xml:lang ) ), ', which means ', count( $gp/t:gloss/@xml:lang ) eq count( distinct-values( $gp/t:gloss/@xml:lang ) ) ). Yes, probably would have been a lot easier to read if it were in XQuery or XSLT with a few variables, but I happened to be in Gizmo with P5 loaded, so this is what I did.

[sydb]

We (@HelenaSabel & I) think that this, like #2826, needs discussion mostly just to decide if constraint goes in p5odds or in P5 proper.

[sydb]

I am assigning to myself to write a constraint (in p5odds.odd unless Council says otherwise) to enforce this.
I am hoping someone else self-assigns to make sure the Stylesheets do the right thing.

BTW, this constraint already exists (in p5odds.odd as "only-one-each-lang").

[martindholmes]

Just flagging up that P5 has three cases of multiple <desc>s with the same language; in all cases, the extra one has type="deprecationInfo", so when writing a constraint we'll need to be aware that this scenario occurs.