(Warning: This post assumes some knowledge of Arabic, although you should be able to follow the argument even without it.)
Everyone knows that Classical Arabic has three short vowels (a, i, u) and three long (ā, ī, ū). But is this true of all varieties of Classical Arabic? Listen to this recitation of sūrat al-'Aʕlā:
There are several distinct Qur'ān recitation traditions, thought to reflect (in part) early dialectal variation in pronunciation. The best known are Ḥafṣ (Asia and Egypt) and Warsh (mainly North and West Africa); the recitation above is in one of the more obscure ones, Khalaf (ʕan Ḥamzah). In it, you will notice that words like šē'a “he willed” شاء, tansē “you forget” تنسى, appear with ē where more common pronunciations of Classical Arabic would use ā. But not all cases of ā are pronounced ē: contrast for instance ġuθā'-an “chaff” غثاء, lā “not” لا. Let's try to figure out what's going on here.
Start with the verbs ending in ā. Verbs which end in ā in the 3rd person masculine singular (“he did”), such as hadā “he guided” هدى, ṣallā “he prayed” صلى, daʕā “he invited” دعا, sajā “it covered with darkness” سجا, divide into two classes in other forms, one ending in y, the other in w: haday-ta “you guided” هديت, ṣallay-ta “you prayed” صليت vs. daʕaw-ta “you invite” دعوت, sajaw-ta “you covered in darkness” سجوت. You have just heard that the former set become hadē, ṣallē. For the latter, we will have to examine different sūras: in the 10th verse of sūrat al-Qamar (1:20) we hear daʕā, and in the 2nd of al-Ḍuħā we hear sajā.
Now ordinary three-letter verbs have the same stem throughout: katab-a “he wrote” كتب vs. katab-ta “you wrote” كتبت. What if the same used to be true of these verbs: *haday-a “he guided” vs. *daʕaw-a “he invited”? (The asterisk means that these are just hypothetical forms.) As it happens, that idea is confirmed if you look at one of Arabic's closest relatives. In Ge'ez, the Semitic classical language of Ethiopia, the cognate verbs are pronounced precisely as reconstructed: with aya (eg ṣallaya “he prayed”) and awa (eg ṣalawa “he roasted”, Arabic ṣalā صلا, ṣalaw-ta صلوت). So if we assume those forms were original, then we can easily see what's going on: in ordinary Classical Arabic both original *aya and *awa end up as ā at the end of a word, but in the Khalaf reading they remain distinct: *aya becomes ē, but *awa becomes ā.
A similar division can be made among verbs with medial ā. Verbs with medial ā in the 3rd person masculine singular, such as zāda “he increased” زاد, ħāqa “he surrounded” حاق, kāna “he was” كان, qāla “he said” قال, divide similarly into two classes in their verbal nouns, one in y, the other in w: zayd زيد, ħayq حيق vs. kawn كون, qawl قول. So we might expect a similar original difference: *zayada, ħayaqa vs. *kawana, *qawala. Sure enough, the pronunciation is as expected. Listen to sūrat al-Baqarah, verses 10 and 11 (about 2:00) and sūrat al-'Anʕām, verse 10 (about 3:00): zēda, ħēqa vs. kāna, qāla. A near-minimal pair is provided by sā'a “he was bad” ساء (sūrat al-Munāfiqūn, v. 2, about 0:50) vs. šē'a “he willed” شاء (already heard in sūrat al-'Aʕlā.)
So – depending on how abstract you are willing to make your representations – this variety of classical Arabic seems to have four long vowel phonemes rather than three. It is also unambiguously more conservative in this respect than the mainstream pronunciation reflected both in the Ḥafṣ reading and in educated standard Arabic, which underscores the philological value of such reading traditions.
(Note: The Qur'ānic Arabic Corpus was useful in preparing this post.)