[OpenType specification] Language system tags for Ancient Greek

[ar-an-ribe]

In the Language System Tags section of the OpenType 1.9.1 specification, both the language system and tag “Greek”/'ELL ' and the language system and tag “Polytonic Greek”/'PGR ' correspond to the ISO 639-2/ISO 639-3 code “ell”, which represents Modern Greek. In ISO 639-2/ISO 639-3, Ancient Greek has a separate code, “grc”.

Should Ancient Greek and Polytonic Ancient Greek share the 'ELL ' and 'PGR ' OpenType language system tags with Greek and Polytonic Greek respectively? Or should Ancient Greek and Polytonic Ancient Greek have their own OpenType language system tags? That is, in the former case, should the Greek and Polytonic Greek lines of the Language System Tags section become

Language system	Language system tag	Corresponding ISO 639 IDs
Greek	'ELL '	ell [Modern], grc [Ancient]
Polytonic Greek	'PGR '	ell [Modern], grc [Ancient]

? Or in the latter case, should those lines become something like

Language system	Language system tag	Corresponding ISO 639 IDs
Greek, Ancient	'GRC '	grc
Greek, Ancient, Polytonic	'PGC '	grc-polyton
Greek, Modern	'ELL '	ell
Greek, Modern, Polytonic	'PGR '	ell-polyton

(assigning two new tags, named 'GRC ' and 'PGC ' here) using the IETF language tags of BCP 47 rather than just their ISO 639-2/ISO 639-3 components?

[tiroj]

I have not identified any typographic need for distinct OTL langsys tags for variants of Greek language, since the differences that exist in the orthographies are addressed at the character level. I have made fonts supporting modern and ancient text, monotonic and polytonic text, Byzantine sigillographic and numismatic transcription, and archaic epigraphy, all using just the grek script tag and the dflt langsys setting.

Distinct OTL langsys tags are only necessary if one needs or wants to make a distinction at the glyph processing level between the typographic presentation of texts. Are there examples of such processing differences that you would expect or desire for the different varieties of Greek identified?

[ar-an-ribe]

Are there examples of such processing differences that you would expect or desire for the different varieties of Greek identified?

The word processor component of LibreOffice supports text of different languages; two of those languages are “Greek” and “Greek, Ancient”. In trying a particular typeface in LibreOffice, which has an OpenType stylistic set that is only available to the 'PGR ' language system tag (this stylistic set changes the shape of the perispomeni diacritic), I found that neither “Greek” text nor “Greek, Ancient” text in LibreOffice could access that stylistic set.

LibreOffice depends upon HarfBuzz for rendering, which in turn uses the IETF language tags of BCP 47 to select the proper OpenType language system tag. Both “Greek” text and “Greek, Ancient” text currently resolve to ISO 639-1 (and thus BCP 47) “el”, which corresponds to ISO 639-2/ISO 639-3 “ell” [Modern Greek]; per the 1.9.1 OpenType specification, “ell” corresponds to the 'ELL ' language system tag. However, “el”/“ell” is not the correct ISO 639 language tag for Ancient Greek; “grc” is its correct language tag, but there is currently no OpenType language system tag that corresponds to the “grc” language tag in ISO 639-2/ISO 639-3 (and BCP 47).

As far as I can tell at this point, for LibreOffice to support both Modern Greek and Ancient Greek, each with monotonic and polytonic variants, it would need to support a couple of extra text languages, presented below in italics:

LibreOffice language	BCP 47 language tag	Current langsys tag	Needed langsys tag
Greek	el	'ELL '	'ELL '
Greek, Ancient	grc	'ELL '	'ELL ' or 'GRC '
Greek, Ancient, Polytonic	grc-polyton	none	'PGR ' or 'PGC '
Greek, Polytonic	el-polyton	none	'PGR '

In addition, HarfBuzz would need to be updated to recognize the “grc” BCP 47 language tag. (It is already able to recognize the “-polyton” subtag.)

The answer to the core question in my initial comment would determine which langsys tag would be needed for the “Greek, Ancient” and “Greek, Ancient, Polytonic” text languages above. It seems to me that the first table in my initial comment would represent the easier solution, i.e. extending 'ELL ' and 'PGR ' in the OpenType specification to represent both “ell” [Modern Greek] and “grc” [Ancient Greek], but I wasn’t sure if there might be a better reason that I hadn’t thought of to choose separate language system tags for both varieties of Ancient Greek.

[tiroj]

In trying a particular typeface in LibreOffice, which has an OpenType stylistic set that is only available to the 'PGR ' language system tag (this stylistic set changes the shape of the perispomeni diacritic), I found that neither “Greek” text nor “Greek, Ancient” text in LibreOffice could access that stylistic set.

Adding additional langsys tags to the spec doesn’t address that, though. It just means there are further tags that may be inadequately supported in fonts.

As I said, I have not encountered any need for language variant langsys tags in Greek fonts. The stylistic set feature you describe can and should have been implemented simply in the dflt langsys entry for the grek script, since the orthographic presence of the perispomeni is a text encoding matter, not a glyph processing one. There is no need for separate langsys entries in fonts to shape Greek text. Indeed, as your LibreOffice example shows, using variant langsys entries for Greek is more likely to introduce problems than to resolve them.

[ar-an-ribe]

In trying a particular typeface in LibreOffice, which has an OpenType stylistic set that is only available to the 'PGR ' language system tag (this stylistic set changes the shape of the perispomeni diacritic), I found that neither “Greek” text nor “Greek, Ancient” text in LibreOffice could access that stylistic set.

Adding additional langsys tags to the spec doesn’t address that, though. It just means there are further tags that may be inadequately supported in fonts.

I agree. That’s why I thought that the first table in my initial comment would be the preferred solution—to preserve the existing langsys tags, but extend their corresponding ISO 639 IDs, since ISO 639 “ell” only represents Modern Greek; the ISO 639 ID for Ancient Greek is “grc”.

Once the OpenType specification associates both of the 'ELL ' and 'PGR ' langsys tags with both “ell” and “grc” ISO 639 IDs, that would allow the HarfBuzz developers to add support for associating BCP 47 “grc” with 'ELL ' and BCP 47 “grc-polyton” with 'PGR '. In turn, the updated HarfBuzz would allow the LibreOffice developers to correct the LibreOffice “Greek, Ancient” language to use BCP 47 “grc” rather than BCP 47 “el” for HarfBuzz rendering, and to add support for polytonic variants of the “Greek” and “Greek, Ancient” LibreOffice languages, using BCP 47 “el-polyton” and “grc-polyton” respectively.

As I said, I have not encountered any need for language variant langsys tags in Greek fonts. The stylistic set feature you describe can and should have been implemented simply in the dflt langsys entry for the grek script, since the orthographic presence of the perispomeni is a text encoding matter, not a glyph processing one.

That stylistic set certainly could have been, and may well should have been, implemented using grek/dflt, but isn’t its implementation using grek/'PGR ' allowed, if not ideal?

In any case, its grek/'PGR ' implementation revealed that:

• LibreOffice isn’t able to render a grek/'PGR ' stylistic set due to not having a text language that uses the corresponding BCP 47 tag that HarfBuzz requires to use the OpenType 'PGR ' langsys tag; and
• LibreOffice’s “Greek, Ancient” language doesn’t use the proper BCP 47 tag/ISO 639 ID for Ancient Greek because HarfBuzz doesn’t support that tag/ID, and HarfBuzz doesn’t support that tag/ID because the OpenType specification doesn’t currently associate that ISO 639 ID with any langsys tag.

Solving the problems in LibreOffice requires solving the problem in HarfBuzz, and solving the problem in HarfBuzz requires updating the OpenType specification to also associate the “grc” ISO 639 ID with both of the 'ELL ' and 'PGR ' langsys tags.