Eugene Oh skrev:
> I'm not too sure about the word-medial status of the viram, but
> word finally it isn't used because the inherent vowel is not
> pronounced at the end of a word. Diachronically speaking this
> is due to apocope in Sanskrit,

Early New Indo-Aryan, actually.  In Old IA/Sanskrit and
Middle IA (Pali, Prakrit etc.) all schwas were still
pronounced, and they still are when reading (or speaking)
these languages, but not when pronouncing loanwords
from these languages into NIA (which practically all
are from Sanskrit, except probably in Jain religious
discourse).  Ironically the viram is written on Skt
loanwords which end in consonants in Skt, but there
it is only a needless adornment.

> but synchronically it is analysed as a "suppression" of the
> vowel, perhaps due to a reanalysis of the inherent vowel as
> epenthetic in all other cases?

Its status is entirely analogous to the _e muet_ in French.
It is even inserted to break up 'hard' consonant clusters
in Skt loans which were clusters in Skt.  E.g. if _mitra_
were borrowed it would become ['mIt@r] although the prescription
is [mItr@].

> In the example you gave, since malmās is a compound, perhaps
> the orthography was simply combined from the two words without
> the complication of modifying the letters. (:

Yes, if it's a 'native' NIA compound.  Clusters which arise in
Skt compounds are written with conjuncts (ligatures), if there
would be a cluster in Skt pronunciation -- virtually only when
the first part of a compound is a consonant stem nominal
(i.e. a stem lacking a theme vowel).