On pg. 125, fn. 2 and fn. 50 I discuss the much-debated question of the length of vowels before final -m in Classical Latin. The conclusion I reach there is that it is very difficult to determine anything about the quantity of these vowels for the Classical period. However, there are two additional arguments bearing on this question that I would like to mention. First, in favor of lenghthening of vowels before final -m Safarewicz 1935 notes that metrical inscriptions sometimes require a long scansion for a vowel before underlying final -m even when that m is omitted in the orthography, e.g. CIL 4.6892:
Quisquis amat nigra nigris carbonibus ardet;
nigra cum video mora libenter aedo.
Whoever loves a dark-skinned girl burns on dark coals.
When I see a dark-skinned girl, I gladly eat blackberries.
(at least that is one possible interpretation; see Varone 2002:57 for other suggestions)
In the pentameter nigra(m) must scan as a spondee and Safarewicz argues that this shows that it was the vowel itself that was bimoraic. But I'm not sure this is conclusive especially since there are also metrical inscriptions which treat the vowel with unindicated -m as short, e.g. CIL 6.1975:
quae tibi crescenti rapuit iuvenile figuram
where the e of iuvenile(m) is the first short of the 5th dactyl. See Fink 1969:448 for more examples.
In favor of a short scansion the always incisive Enrico Campanile 1973 made two observations. First in Latin loanwords into Brittonic sequences of final -um do not behave like -ō or -ū in that they do not trigger vowel affection of the preceding vowel. Second, it is a well established fact that Latin hexametric poets of the Classical period avoided elision of cretic sequences. The exceptions in Vergil involve final -ō in 3rd declension n-stem nominative singulars (Ecl. 3.84 Pollio amat nostrum quamvis est rustica Musam) and the 1st sg. ending -ō (Aen. 11.503 Audeo et Aeneadum promitto occurrere turmae). But these exceptions are only apparent since it is very probable that these vowels were for Vergil synchroncically short as a result of so-called cretic shortening. When it comes to vowels before final -m in words whose preceding two syllables form a trochee, Vergil has no compunction about eliding the final syllable, e.g. Aen. 2.667 Alterum in alterius mactatos sanguine cernam; Aen. 4.387 Audiam et haec manis veniet mihi fama sub imos. This suggests that these were not cretic sequences and therefore that the vowel was short. I am inclined to agree with Campanile especially since this evidence agrees with Priscian's statement, and the fact that apices are never used on a vowel before final -m in inscriptions which use apices correctly.
Campanile, Enrico. 1973. Sulla quantità della vocale che precede -m in latino. Italia dialettale 36:1–6.
Fink, R. O. 1969. A long vowel befor final -m in Latin? American Journal of Philology 90:444–52.
Safarewicz, Jan. 1935. Les voyelles nasales en latin. In Atti del III congresso internazionale dei linguisti (Roma, 1933). Florence 176–9.
Varone, Antonio, 2002. Erotica pompeiana. Love inscriptions on the walls of Pompeii. Rome: Bretschneider.