Simplify grammar for implementing it in lutaml-model

[andrew2net]

1. An issue with Inheritance can be solved with a workaround, but if some element inherits from a definition that inherits from another definition, it becomes a problem. For example, FormattedString inherits from LocalizedStringOrXsAny, and relation/description inherits from FormattedString.
2. Shale can't handle choice/group cases.
3. ...

[opoudjis]

We're going to get rid of FormattedString anyway, since we're using Basicdoc for all formatted text.

[opoudjis]

And we will get rid of variant.

So

LocalizedString1 =
  # multiple languages and scripts possible: comma delimit them if so
  LocalizedStringAttrs,
  text

LocalizedString =
  LocalizedString1 |
  element variant { LocalizedString1 }+

LocalizedMarkedUpString1 =
  # multiple languages and scripts possible: comma delimit them if so
  LocalizedStringAttrs,
  TextElement+

LocalizedMarkedUpString =
  LocalizedMarkedUpString1 |
  element variant { LocalizedMarkedUpString1 }+
  
# Unlike UML, change type to format: type is overloaded
# Would be need if plain were default value and could omit the attribute
# Added LocalizedStringOrXsAny
FormattedString = 
  # attribute format { ( "plain" | "html" | "docbook" | "tei" | "asciidoc" | "markdown" ) }?,
  attribute format { ( "text/plain" | "text/html" | "application/docbook+xml" |
    "application/tei+xml" | "text/x-asciidoc" | "text/markdown" | "application/x-metanorma+xml" | text ) }?,
  LocalizedStringOrXsAny

LocalizedStringOrXsAny1 =
  # multiple languages and scripts possible: comma delimit them if so
  LocalizedStringAttrs,
  ( text | AnyElement )+

LocalizedStringOrXsAny =
  LocalizedStringOrXsAny1 |
  element variant { LocalizedStringOrXsAny1 }+

becomes

LocalizedString =
  # multiple languages and scripts possible: comma delimit them if so
  LocalizedStringAttrs,
  text

LocalizedMarkedUpString =
  # multiple languages and scripts possible: comma delimit them if so
  LocalizedStringAttrs,
  TextElement+

LocalizedStringOrXsAny =
  # multiple languages and scripts possible: comma delimit them if so
  LocalizedStringAttrs,
  ( text | AnyElement )+

[opoudjis]

choice/group:

... ugh, unhappy about simplifying these, but:

FullNameType =
    name_abbreviation?,
    (( prefix*, forename*, formatted-initials?, surname, addition* ) | completeName ),
    biblionote*, variantname*

becomes

FullNameType =
    name_abbreviation?,
    prefix*, forename*, formatted-initials?, surname?, addition*, completeName?,
    biblionote*, variantname*

and

address =
  element address {
    # iso191606 TODO
    (street*, city, state?, country, postcode? ) | formattedAddress
}

becomes

address =
  element address {
    # iso191606 TODO
    street*, city?, state?, country?, postcode?, formattedAddress?
}

[opoudjis]

the complexities are:

• choice
• more then 1 level inheritance
• recursion

[opoudjis]

Also getting rid of Name Variants:

orgname = element name { LocalizedString | NameWithVariants }

NameWithVariants =
  element primary { LocalizedString },
  element variant { NameWithVariants1 }*

NameWithVariants1 =
  attribute type { text },
  element primary { LocalizedString },
  element variant { NameWithVariants1 }*

This is recursive and complicates things unnecessarily, given we have multiple name strings. This hasn't been used anyway in Relaton. We will just have optional type on names.

[opoudjis]

Changing bplace from choice/group

bplace = element place {
  text | ( bibliocity, biblioregion*, bibliocountry*)
}

to follow other elements in having an optional formattedplace element, which concatenates all the elements into a single element:

bplace = element place {
  bibliocity?, biblioregion*, bibliocountry*, formattedPlace?
}

cf. formattedAddress

This is a breaking change; so <place>Richmond, VA</place> becomes <place><formattedPlace>Richmond, VA</formattedPlace></place>, as an alternative to <place><city>Richmond</city><region>VA</region></place>

[opoudjis]

Also flattening the choice/group in keyword,

bkeyword = element keyword {
    LocalizedString |
    (
       element vocab { LocalizedString },
       vocabid+
    ) |
    (
       element taxon { LocalizedString }+,
       vocabid+
    )
}

to

bkeyword = element keyword {
   element vocab { LocalizedString }?,
   element taxon { LocalizedString }+, 
   vocabid*
}

Another breaking change: if the keyword is just text (LocalizedString), it will now be encoded as vocab

[opoudjis]

There is an instance of recursion in the grammar that I have introduced somewhat recently, and that I really do not want to get rid of:

organization = element organization { OrganizationType}

OrganizationType =
    orgname+, subdivision*, abbreviation?, uri*, org-identifier*, contact*, logo?

orgname = element name {
   attribute type { text }?,
   LocalizedString
}

subdivision = element subdivision { 
   attribute type { text }?,
   OrganizationType
}

That is, model organisation subdivisions as full organisations themselves; this proved necessary for JIS, as subdivisions had internal structure and could not be restricted to just name strings.

[opoudjis]

I'm flattening LocalizedStringAttrs into other type definitions, because this:

LocalizedStringAttrs =
  # multiple languages and scripts possible: comma delimit them if so
  attribute language { text }?,
  attribute locale { text }?,
  attribute script { text }?

LocalizedMarkedUpString =
  # multiple languages and scripts possible: comma delimit them if so
  LocalizedStringAttrs,
  TextElement+

roledescription =
  element description { LocalizedMarkedUpString }

is supposedly beyond the ability of Shale to deal with, as two levels of indirection.

That Shale apparently cannot deal with

A = attribute B

C = {
  B
  text
}

D = { C }

is reprehensible, and I deserve an apology for it, @ronaldtse . I'm sure I won't get one.

[opoudjis]

I've simplified the Relaton grammar as far as I'm prepared to. Over to you, @andrew2net

[opoudjis]

In preparation, getting rid of variant as used within Metanorma

[ronaldtse]

There is an instance of recursion in the grammar that I have introduced somewhat recently, and that I really do not want to get rid of:

organization = element organization { OrganizationType}

OrganizationType =
    orgname+, subdivision*, abbreviation?, uri*, org-identifier*, contact*, logo?

orgname = element name {
   attribute type { text }?,
   LocalizedString
}

subdivision = element subdivision { 
   attribute type { text }?,
   OrganizationType
}

That is, model organisation subdivisions as full organisations themselves; this proved necessary for JIS, as subdivisions had internal structure and could not be restricted to just name strings.

This kind of recursion can always be modeled through relationships, ie. a flat structure for organizations then link them together using attributes or elements.

[opoudjis]

Org 1, Org 2, Org 3, Org 1 is child of Org 2?

... Ugly. I'll wait on what @andrew2net has to say.

[opoudjis]

Make PureTextElement recursive, so that its instances of em, strong do not invoke instances outside of PureTextElement; and use PureTextElement more in Relaton grammar, where more full TextElement support is unneeded. @andrew2net will treat the full prose elements like Abstract, which do need full TextElement support, as Nokogiri XML rather than try to deal with them in Shale.

[opoudjis]

In biblio.rnc, TextElement* is now restricted to abstract and formattedref. Everywhere else, including titles, is constrained to PureTextElement*