And here are my opinions on these questions:
p5subset.xml has all the languages (as it does); customization1.odd specifies >@docLang="en" customization2.odd, which points to customization1.odd, specifies >@docLang="it" What language should the output schema use?
Schemas compiled against customization1 will see help popups in English, and those compiled against customization2 will see them in Italian. That is, assuming neither ODD has done anything to remove the relevant components. The languages available from a compiled ODD should presumably be managed in the same way as anything else. So if I delete an English <gloss> in my customization, and then compile that to use as a basis for a further customization, the compiled version will no longer have an English gloss to use, so specifying @docLang='en' won't produce anything. In general specifying @docLang only expresses an output preference; it has nothing to do with determining what components are available from a compiled version of the ODD. 2. Imagine:
customization1.odd modifies an element, replacing the English gloss with >something different. customization2.off specifies @docLang="es" Where should the >gloss be taken from in the output of the chain?
I think I already answered that. If "customisation2.off" is using a compiled version of customization1.odd then the spanish gloss will be taken from that, if it's there.
3. Should it be part of the system that you can specify a series of languages in >order of preference, from which text will be taken if something is not available >in the main docLang?
Seems like an over-complication. There is a canonical language in which ODD glosses are written, which is currently labelled "en" (unless and until the TEI becomes a purely Chinese -- say -- project). So it is not unreasonable to say if @docLang requests a language that's not available (for whatever reason), a version using the canonical language is the fallback. If you really want to I suppose you could supply multiple values for @docLang to indicate acceptable languages, and then choose whichever comes first on the list, but this seems like over-egging the pudding to me.