Etymology of 宵【よい】

This is a loose follow-up of my first question here on JLSE. I tried to check gogen-allguide.com and 大辞林, but couldn't find anything definitive. Unfortunately I don't know any other resources for checking etymologies for (native) Japanese words.

So, my question is: What is the etymology of 宵【よい】? My guess would be 夜(よ)日(ひ).

Answer

Gogen Allguide and my dead-tree version of Sh­ōgakukan Kokugo Dai Jiten in JA mention that, in the Heian period and earlier, the terms for "evening, night" were somewhat specific, going in order:

• ゆうべ
  Early evening


• よい
  Late evening, early night: dusk


• よなか
  Night


• あかとき
  Pre-dawn


• あけぼの
  
  Dawn

I can't find anything definitive at present about the derivation of 宵. I do note, however, that 夜日 is a separate word in its own right, yohi: "night and day, day and night". So that's probably not the source of 宵.

I do see that the older kana spelling was よひ. I note also that 夕 yū of similar meaning has an older kana spelling of ゆふ. I wonder then if these two might be cognate, with よひ representing some kind of inflection. Or, possibly, よひ might be the noun stem form of a hypothetical verb よふ, consisting of 夜 "night" + ふ "suffix indicating iterative or continuous action, or ongoing change". That said, I can't find evidence of any such verb in my sources to hand.

---

Update.

The link to JSTOR that snailboat added above in a comment previously didn't work for me, possibly due to net congestion or server trouble. I have since read through that paper by J. Marshall Unger, New Etymologies for Some Japanese Time-Words. In it, he provides the following theory on page 40 for the derivation of modern 宵{よい}, ancient yopi (my additions in [square brackets] for clarity):

Because free native nouns in Old Japanese could begin with p but not r, the derivation of yopi 'evening twilight' is not as problematic as that of yoru, though by no means certain. It was probably a relatively late reduction of a whole phrase such as *yo pikari 'night shining', with the same *yo 'night' morpheme. According to pJ [proto-Japanese] non-final mid-vowel raising, an ancient compound of the form *yo-pi would have yielded OJ [Old Japanese] *yupi, not yopi, and OJ pi 'sun, day' was not used for 'light'.

I'm not sure what pJ non-final mid-vowel raising is beyond the summary provided by Unger (apparently outlined in more detail in a separate paper by Frellesvig and Whitman in 2004), but from the context of Unger's paper, I find myself unconvinced—there are OJ terms that don't seem to exhibit this purported yo- > yu- shift. I'm also unconvinced that pikari would shorten to just pi, given that in all other contexts that I'm aware of, the shortest form is pika, though I am certainly open to some other term being the cognate source for the -pi in yopi.