on Latin -a non-feminines
> what are the exceptions ? nouns like agricola are not derived from a
> corresponding form in -us are they ?
I worded that badly. Apart from masculines in -a, such as agricola, there
are forms like aquila (f) used for male eagles, and incola, advena, auriga
which are of "common" gender, meaning they take the form of the adjective
appropriate to the natural gender of the person to whom they are applied.
It may be the case that masculines like agricola, scriba and so on
originally had the ending -a:s, as in Greek. (A very few show this).
> What are the current ideas on exactly how/why this -h2e form came to be
> used as the adjectival ending in PIE -os adjectives
The literature has only wild guesses (at least in my opinion) but some are
more believable than others. The best of these guesses suggest that at an
early stage in its development PIE may have been ergative (or something of
that kind - it is hotly debated). This means that nouns and verbs
distinguished between active and non-active ("stative") vocabulary items.
This explains that many apparent doublets for words like fire, water, and so
on. Nouns and adjectives therefore had to be marked for active or stative
(or whatever the different categories were) if they were not already so
marked or understood. Hence you get the origins of agreement.
If this seems unpersuasive, it's because I have missed out the details. It
does seem likely that PIE originally had an animate/inanimate distinction of
some kind, and that suffixes such as -h2e which end up as a feminine marker
began life with a different meaning, now lost.
That probably tantalises more than it explains, but I hope it does at least
suggest the process.
Peter