Document accursedUnutterablePerformIO properly about bytestring HOT 3 CLOSED

Mmm. My intention of "documenting" it was to encourage people not to use it (I'd seen people use it without realising). To give people a proper understanding of it reqires quite a bit of effort. In particular I don't have a complete explanation of when it is safe. I'd consider a PR though.

As for deprecated, no it's not deprecated. The export list says:

        -- * Deprecated and unmentionable
        accursedUnutterablePerformIO, -- :: IO a -> a
        inlinePerformIO               -- :: IO a -> a

because in this section is one thing that's deprecated and one thing that's unmentionable, not that both are both. Could change that 'and' to an 'or' :-)

from bytestring.

My intention of "documenting" it was to encourage people not to use it

I know, but I think that's a dirty trick. I would help but have no idea how it's safe to use. For now I'll just try to use it strictly as you do in bytestring where I can see a performance benefit.

from bytestring.

@jberryman I think it's safe if the effects are completely read only and can be duplicated and run concurrently. So yes, reading from a foreign ptr when that foreign ptr memory region is guaranteed to be immutable is ok I think. Examples of "very bad things" include allocating memory or any other resource (those are what some of the linked examples did). Where the dividing line is exactly is very hard to pin down.

from bytestring.