Motivations from this issue from Sporth.

dmetro has a few shortcomings which should be specified in the documentation.

Mainly, the things I wish to bring up:

• sample accuracy vs sample precision (and how the latter is more important in this context)
• why dmetro adds one sample to the duration in the seconds to

I found that mel $ repeat ... does not work (it is type-correct, but it gives stack overflow)

mix $ sco (\ f -> return $ osc $ sig f ) $ mel $ repeat $ mel [ temp 300, rest 1 ]

On reason is implementation (mel uses foldl1, at least the default implementation does, https://github.com/spell-music/temporal-media/blob/master/src/Temporal/Class.hs#L24 ) but if I use

... foldr1 (