In (Ancient) Quenya, unstressed medial long vowels shortened, a change mentioned in the Outline of Phonetic Development (OP1) from the 1930s:

(2) After this syncope, but still before the Q. accent shift: Long vowels that remained in unstressed medial syllables were shortened, but not lost. So Òrōmḗ > Oromḗ (Q. Orome) (PE19/58).

As this quote indicates, this shortening took place after the Quenya syncope (because the long vowel prevented the syncope), but before the Quenya stress shift. Indeed, there are examples of long vowels both shortening and not shortening, and this must be the result of variations in primitive stress patterns. Here are examples of shortening:

• ᴹ✶álākō > ᴹQ. alako (Ety/ÁLAK).
• ✶arātā̆ > Q. arata (PE17/49, 186).
• ᴹ✶narāka > ᴹQ. naraka (Ety/NÁRAK).
• ✶Orǭmē or Arǭmēz > Q. Oromë (PE17/153, WJ/400).
• ᴹ✶terēn(ē) > ᴹQ. terene (Ety/TER).

Here are examples where shortening does not occur:

• Q. aráto “champion” (SA/ar(a)).
• Q. imíca “among” (VT43/30).
• Q. ilúvë “allness” (WJ/402, Ety/IL).
• ᴹQ. naháme “summons” (EtyAC/KHAM²).
• ᴹQ. palúre “surface, bosom (of Earth)” (Ety/PAL).

Note that aráto has an alternate form arato derived from a primitive form where the long vowel was unstressed: ✶árātō (PE17/118), which is a strong indication that shortening (or lack thereof) was tied to stress. There are, however, a few aberrant examples where shortening occurs despite the stress mark:

• ᴹ✶terḗwā > ᴹQ. tereva “fine, acute” (Ety/TER; EtyAC/TER).
• ᴹ✶ontā́ro > ᴹQ. ontaro “parent” (Ety/ONO).

Indeed, Tolkien seems to have vacillated on the interaction of the Quenya stress shift and the Quenya syncope, and these vacillations may have influenced the nature of unstressed long-vowel shortening as well. However, some of these apparent “vacillations” might be the result of grammatical reformations. For example:

In general such interior long vowels as had remained at end of archaic period were still kept - since the accent in fact so frequently stood on them, as in torṓmā > toróma. But long vowels in open syllables that were now in the antepenult and just before the main stress tended to be shortened in PQ and though spelling is usually maintained are always shortened in TQ. But a stress-less long vowel is usually maintained in the stem-syllable. Thus torṓmànna > PQ torō̆mánna > TQ torománna. But kṓmā, kṓmallṑ > Q. cṓma, cōmállō̆ (OP1, PE19/59).

The treatment of medial long vowels was however complicated by many reformations. Phonetically oxytone nouns ending in a long vowel would retract [in] forms uninflected or followed by an asyllabic inflection (so Ulmṓ, Ulmṓn > Úlmō, Úlmōn) but the earlier accent would remain before a syllabic inflection, gen. Ulmṓvā. But in declensions the oxytone type became a special form limited to animates. Hence kiryā́, kiryā́va > kíryā, kíryăvă after type kándā, kándāvā̀ > kandava, aided also by fact that the two coalesced in longer inflected forms as kiryālī́nen, kandālī́nen > kiryalīnen, kandalīnen (OP1, PE19/60).

These various treatments of stressed and unstressed medial long vowels probably influenced the development of prosodic lengthening in Quenya.

Conceptual Development: Tolkien mentioned similar behavior for Early Qenya in The Qenya Phonology from the 1910s:

The practical results were a throwing forward of the accent, so that: ... (4) Vowels still preserved long which were by this change unaccented between 2 accents were shortened (PE12/5).

Under the Qenya system accent came therefore under the influence of syllable length or weight but where rhythm clashed with this the unaccented syllables were as far as possible shortened and lightened (PE12/6).

In Early Qenya, it seems the shortening occurred after the stress shift, not before. As a result, in Early Quenya it seems the connection between stress and vowel length was even stronger:

Therefore vowel length is no longer etymological and morphological purely, as it was in Eldarin and still largely remained in Cor-eldarin, but is mostly to be determined by that vowel’s place in the word and its neighbourhood. Length marks are therefore given up: a stress mark is retained instead (PE12/6).

And:

Only vowels under a stress could be long (PE12/6).

In other words in Early Quenya an unstressed vowel is always short, and any long vowel in an open syllable (one followed by a single consonant) is almost always stressed. This was no need to distinguish length markers and stress markers in the Early Qenya of the 1910s because a long vowel was generally also where the stress fell. This was frequently true in later conceptual iterations of Quenya, but not invariably so. Compare the ablative form cōmállō̆ of cṓma [“from the ball”?] against the allative form torŏmánna of torṓma [meaning unknown], but this vowel-shortening within inflected forms was reflected only in pronunciation, not spelling, as noted in the quote given above (PE19/59).