I would like to address some points of Paul Strack's article on the Quenya prosodic lengthening.
Another interesting set of examples come from the genitive partitive-plural suffix -lion, which is a disyllabic suffix that also triggers prosodic lengthening: vanimálion “of beautiful ones” (LotR/981), malinornélion “of yellow trees” (LotR/467; PE17/80). Similar behavior is seen in the genitive plural suffix -ion: Atanatárion “of the Fathers of Men” (MR/373)
Note that the word malinornélion breaks the rule that the lengthening cannot occur immediately after the main stress of the original word stem. Perhaps “long suffixes” like this are an exception because the originally stressed syllable is now far enough back that it can no longer hold the main stress. If so, a more accurate description of the conditions where prosodic lengthening occurs would be “after suffixion the word ends in three light syllable” (but this is the only example where this alternate rule would be required).
Vanimálion, malinornélion, atanatárion are not vanima/malinorne/atanatar + lion/ion. They are the genitives of vanimáli/malinorneli/atanatári, so vanimáli/malinorneli/atanatári + on. No disyllabic suffix, just an ordinary monosyllabic one.
Now, about malinornélion alone. By the basic regular process, there would be mălĭnŏ́rnĕlĭ > *mălĭnŏrnĕlī́ŏn; however, the sequence long vowel - hiatus - another vowel is not permitted in Q. (or to be precise, it is, but only under certain conditions that are never met here), so Paul correctly says that the stress must retract to the preceding -e-.
But what about the length of this vowel then? Well, by my count, there are 8 attested word forms of the exact same pattern as malinorneli+on:
3 of them (aldali > aldalion, Orome > Oromeo, almare > Almarian) show no lengthening of the stressed vowel; on the other hand, 2 of them (almare > Almáriel, poldore > poldórea) do show it; and another 2 are attested both with an without it (malinorneli > malinornélion/malinornelion, almare > almárea/almarea). The last one (sindie > sindieo) is indecisive, as another hiatus prevents its antepenult from lengthening anyway.
So I'd say this demonstrates fairly well that in this particular situation (stress moved one syllable to the right instead of two because of the hiatus before the newly added syllable), the lengthening of the newly stressed syllable is regular, but optional. May take place, but doesn't have to.
Prosodic lengthening is also inhibited if the base word is “too short”. If the base word has only two syllables
The possessive suffix is a bit of a special case: there examples of this suffix triggering prosodic lengthening even in cases of disyllables: Ulmóva (PE19/60; PE21/80), taríva (PE17/76; PE21/14), huinéva (Ety/PHUY). This is not a pattern Tolkien followed consistently, however: forms like Eldava and lasseva also appear.
As a matter of fact, -va is not the only suffix that triggers the stress shift and consequent lengthening in disyllables. There is also kasar > kasári, lassi > lassínen and maybe more. I would actually say that this is a case similar to the above: a syllable added to a light final syllable of a disyllabic word allows for optional shift and lengthening.
And a third class of words where this optional shift&lengthening seems to take place are disyllables with another disyllabic element added, if the result is a sequence of three light syllables: anar > Anárion, alta > Altáriel, alar > Aláriel, possibly *eled > Elériel; but beside these also atan > atanatar, Elda > Eldalie, and even alta > Altariel.
Now, what have all these three patterns in common is that the stress is in each of them moved one syllable to the right. So in fact, a simple set of additional rules could be postulated:
• If the stress is moved to a light syllable two syllables to the right, the newly stressed syllable is mandatorily lengthened.
• If the stress is moved to a light syllable one syllable to the right, the newly stressed syllable is optionally lengthened.
• If the stress is either left where it was or moved to a heavy syllable, no lengthening (naturally) takes place.
there are definitely cases of prosodic lengthening where no long vowel existed primitively, such as Atanatári (S/103) and Ilúvatáro (PE21/83)
Actually, these may have primitively been made out of the subjective case form -atār (cf. PE 21/76). I was unable to find a clear, indisputable case of a primitive short vowel prosodically lengthened (apart from the etymologically late analogical *karúvane* of PE 22/105, discussed in Paul's article separately).
I suspect the stress pattern would still be sìndiéo even if it is not reflected in a lengthened vowel
Why? This -ĭĕŏ is three light syllables with a hiatus before the suffix, so there seems to be no reason to doubt that the accentuation is the regular -ĭ́ĕŏ.
caraitië “activity” [karáitie].
caraitiéli “some activities” [kàraitiḗli].
caraitielínen “by means of some activities” [kàraitielī́nen].
Again, the instrumental of caraitiéli must be caraitiélinen. No reason to be otherwise.
There are some cases where prosodic lengthening is expected but does not occur. Some of the examples that show lengthening also appear without it: malinornelion, vanimalion and Ilúvataro are all attested (PE17/80; PE17/111; S/322). There are also some compound forms that should lengthen but don’t, such as Tasarinan “Willow-vale” = tasari + nan (LotR/469)
As said above, malinornelion is rather regular. Ilúvataro is a slip on Christopher Tolkien's part (his father's original on PE 21/83 has correct Ilúvatáro). And tasarinan is faulty only if analysed as the instrumental of tasare; if from (also attested) tasar, it "becomes" perfectly OK. So of these only vanimalion seems truly wrong (probably a mere slip, considering that it's a hand-made note, while the printed and published form is regular vanimálion).
But there are two more examples that break the rules: Anárioni and Isildurioni. They're supposed to end in lengthened -óni.