It occurred to me that the Lith. o-stem ins.pl. -ai~s cannot
go back to the usually reconstructed PIE form *-o:is (or
*-ôis). That would have given -uis, since uo~R gets reduced
to -uR (e.g. Gpl. -uN, Dsg. -ui) and uó gets reduced to ù
(I.sg. -ù). The Lithuanian form must go back to *-ois.
Slavic -y is inconclusive, as -VRs gets lengthened to -V:Rs,
so *-ois > *-uih > *-u:ih > -y, the same as *-o:is or *-ôis.
The other form pointing to a long vowel, Skt. -aih. (i.e.
-a:is], would seem to be inconclusive too, as Sanskrit has,
independently of Slavic (and independently of
Szemerényi-lengthening in pre-PIE) the same soundlaw too:
e.g. acc.pl. of i- and u-stems -i:n/-i:s, -u:n/-u:s, like
Slavic -i, -y, < *-ins, *-uns.
So, unless I'm missing something (Greek -ois, Latin -i:s are
inconclusive too), the ending should be reconstructed as
*-ois, based on the conclusive Lithuanian evidence.
=======================
Miguel Carrasquer Vidal
mcv@...