After short vowels vanished before morpheme boundaries in compounds, where a long vowel appeared before any resulting consonant clusters that vowel was usually shortened. This is analogous to the same change that occurred in Primitive Elvish, and in all Eldarin languages in general, which mostly did not tolerate long vowels before consonant clusters. The clearest examples of this sound change are in Sindarin, but this phonetic development seems to apply at all conceptual stages:

• ✶mēlā̆-mbar > mîl-mbar > S. milbar “dear home” (PE17/164).
• ✶nēn-talma [> *nīn-dalmh] > S. Nindalf “Wetwang” (PE17/167).
• ✶gōr(i)kubā [> *gūr-gof] > S. gurgof “traitor” (PE22/155).
• †Rochír-rim > S. Rohirrim “Horse-lords” (UT/319).

At this stage of phonetic development in Sindarin and Noldorin, the primitive long vowels [ē], [ō] and [ā] no longer existed, since [ē], [ō] became [ī], [ū], whereas [ā] became [ǭ] and then [au]. However, a new long vowel [ȳ] had appeared (in Sindarin but not Noldorin) as a phonetic development from [eu], [iu] and [ju]. It seems that [ȳ] shortened before clusters but not before single consonants:

• ✶julmā [> *ȳlmh] > S. ylf “drinking-vessel” (WJ/416).
• S. cýra “renew” and S. cýron “new-moon” (VT48/7).
• S. dýgar “mistaken act; doing a wrong thing” (PE17/151).
• ✶jūneke [> *ȳneg(a)] > S. ýneg “twelve” (VT47/41), although this example appeared in other places as yneg (VT48/8).

The shortening of [ī], [ū] before clusters is complicated by the fact that [ī], [ū] often shortened in polysyllables anyway, even before single consonants. There are also a fair number of examples where long [ī] did not shorten before clusters:

• S. Círdan “Shipwright” (LotR/240), sometimes appearing as Cirdan (WJ/8; WR/76).
• S. Dírhael “*Wise Man” (LotR/1057).
• S. íðra “long for, desire” (PE17/112).
• S. Mírdain “Jewel Smiths” (S/286).
• N. nírnaeth “lamentation” (Ety/NEI) vs. nirnaeth.

It’s difficult to extrapolate a pattern from such a small number of words. Many of these long vowels appear before r, as noted by David Salo (GS/§4.130), but compare this to S. gurgof and Rohirrim above.

For a discussion of the possible shortening of [ǭ], see then entry on how [ǭ] became [au].