Re: Proposal: merge either into transformers

On Mon, Apr 28, 2014 at 4:15 PM, Ross Paterson wrote:

On Sun, Apr 27, 2014 at 10:47:56AM +0300, Michael Snoyman wrote:

...

On Sat, Apr 26, 2014 at 9:57 PM, Ross Paterson wrote:

...

would be incompatible with the lazy WriterT in subtle ways

That would be troubling, but I'm not sure in which ways it's incompatible. Do you have any examples?

The Applicative and Alternative instances would have different contexts, and there would be no instances for Foldable, Traversable, Eq, Ord, Read or Show.

If we have deprecation of the module in its entirety on the table, I think it's acceptable to consider dropping some instances. However, I don't see Eq, Ord, Read, or Show instances for strict WriterT in transformers 0.3. Applicative seems like it should be identical in behavior to what we have right now. I'm not completely certain, but it seems the same is true for Alternative.

...

The advantage of having WriterT implemented in terms of strict StateT is that many people will automatically get the fix when upgrading to transformers 0.4. Also, the writer API itself is very convenient for many common use cases, so it would be nice if there was a version available that didn't leak memory.

Invisibly increasing strictness cuts both ways: some people get space savings, while others get non-termination. (Although one might say they're not using it properly, that seems a harsh punishment.) But I'm more concerned about the complication. Currently the transformers are transparent, and people can think about the type structure of the monad they're building. This very transformer is already there; it seems simpler to direct people to it instead of having another copy. The API point is an issue, though.

Here's the question I'd ask, which I honestly don't know the answer to. We have three proposed WriterT implementations: lazy, current-strict, and state-strict. We have two conflicting desires: program termination and space savings. We know there are cases where lazy allows termination where state-strict does not. We know there are cases where state-strict allows space savings where neither lazy nor current-strict do. The question is: are there cases where current-strict: 1. Gives space savings that lazy does not? 2. Gives termination where state-strict does not? I *think* the answers to these questions are "no" and "yes", meaning that current-strict in its current form can *always* be replaced by lazy, without losing anything. If that's the case, I'd say this is a very simple transition in 0.4. Note that we've been discussing all of this in terms of WriterT, but the same exact discussion should apply to RWST. Michael