To tie the knot I summarised this as a blog post:
http://neilmitchell.blogspot.co.uk/2016/07/why-did-stack-stop-using-shake.html

Thanks, Neil


On Mon, Jul 11, 2016 at 8:24 PM, Neil Mitchell <[email protected]> wrote:
> Very interesting comments from all.
>
> In my opinion, Shake excels at a few domains.
>
> * If you have lots of complex interdependencies, Shake lets you manage
> them nicely. That's not really the case here, but is in large
> heterogeneous build systems, e.g. the GHC build system.
>
> * If you are writing things quickly, Shake lets you manage
> exceptions/retries/robustness quickly. For a project which has the
> effort invested that Stack does, that's less important, but for things
> like MinGHC (something Stack killed), it was critically important
> because no one cared enough to do all this nasty engineering.
>
> * If you are experimenting, Shake provides a lot of pieces (resources,
> parallelism, storage) that help explore the problem space without
> having to do lots of work at each iteration. That might mean Shake is
> more of a benefit at the start of a project than in a mature project.
>
> FWIW, the concurrent execution stuff in the middle of Stack actually
> looks fairly hairy! I see a dependency based build system, with
> features like staunch mode and parallelism. It's almost certainly an
> O(n^2) algorithm in several places, although I suspect n is typically
> small enough. I guess it assumes topological sorting and no dead links
> - but I would assume all those are guaranteed in advance. esFinalLock
> gives me a gentle shiver. esInAction looks like a space leak, unless
> forced by actionDo. ActionContext worries me a bit, is it meant to
> list all actions that aren't this one? That seems a very weird
> argument.
>
> All those comments about Execute.hs are not intended to persuade you
> to move to Shake, but I do see plenty of resemblances of Shake in
> there. In many ways you have written a mini-Shake.
>
> Thanks, Neil
>
> On Mon, Jul 11, 2016 at 6:27 PM, Edward Z. Yang <[email protected]> wrote:
>> No Dan, I think you're exactly right.
>>
>> I have a similar feeling when working with the cabal-install codebase.
>> Without an API like Shake, you manually construct and run the graph of
>> operations you want to do.  Maybe it is simple, but there are so many
>> other things which you could be caching / parallelizing more aggressively.
>>
>> Edward
>>
>> Excerpts from Dan Burton's message of 2016-07-11 10:20:13 -0700:
>>> As I understand, shake is good at untangling "I need this to do that, I
>>> need that to do the other" sorts of tasks. Stack has a lot of these things,
>>> e.g. "I need to read the stack.yaml to...", "I need to download the
>>> resolver build plan from stackage.org to...", and so forth. A deep dive
>>> into using shake pervasively may prove beneficial, although I have
>>> practically no experience with shake so I could be completely off base here.
>>>
>>> -- Dan Burton
>>>
>>> On Mon, Jul 11, 2016 at 9:55 AM, Michael Sloan <[email protected]> wrote:
>>>
>>> > Yeah, one thing I've wanted shake integration for is providing is
>>> > collection and display of performance stats - a coarse impression of where
>>> > time is spent.
>>> >
>>> > Thing is, we really aren't dealing with a make-like build system involving
>>> > a variety of different interdependent tasks.  This is shake's killer app /
>>> > intended domain.  Instead, we've just got some cabal builds to do.  The
>>> > implementation of concurrently executing such tasks is very simple:
>>> > https://github.com/commercialhaskell/stack/blob/master/src/Control/Concurrent/Execute.hs
>>> >
>>> > So I'm also not sure if it's the right decision, but I'm not sure how much
>>> > it matters.  Stack is a lot more than just the "do build steps" bit, 
>>> > though
>>> > that is certainly the core.
>>> >
>>> > It could be interesting to consider how difficult it would be to switch
>>> > some choice bits over to shake, but I'd want to see a couple convincing
>>> > advantages, and I'd hope for the code impact to be low.  Reuse of the
>>> > performance stats stuff is one potential advantage.
>>> >
>>> > -Michael
>>> >
>>> > On Sat, Jul 9, 2016 at 10:56 AM, Michael Snoyman <[email protected]>
>>> > wrote:
>>> >
>>> >> It wasn't made for deep reasons, it was very likely done for bad reasons
>>> >> driven by Chris and my lack of experience with Shake when working on the
>>> >> initial Stack code. We had a lot of trouble getting reliably coordination
>>> >> between Stack's, Cabal's, and (to a lesser extent) GHC's dependency
>>> >> tracking, and adding in an extra layer we didn't have full understanding 
>>> >> of
>>> >> made it worse. I have no idea if using Shake would have been better or
>>> >> worse in the end, but doing it manually - a way Chris and I had a lot of
>>> >> experience with - was the less risky path.
>>> >>
>>> >> I _will_ say that on other customer projects, we use Shake extensively,
>>> >> so don't take the lack of usage in Stack as an aversion to Shake. It was
>>> >> just a pragmatic decision when we were trying to churn out a first 
>>> >> version
>>> >> quickly.
>>> >>
>>> >> On Sat, Jul 9, 2016 at 3:00 AM, Edward Yang <[email protected]> wrote:
>>> >>
>>> >>> This is not meant as a leading question, I am just curious and would
>>> >>> like to know more.
>>> >>>
>>> >>> According to
>>> >>> https://github.com/commercialhaskell/stack/wiki/Stack's-origins Stack
>>> >>> used to use Shake. Looking at the dependencies of Stack, it no longer 
>>> >>> does
>>> >>> so. What caused you to rebuild the functionality of Stack into Shake?
>>> >>>
>>> >>> --
>>> >>> You received this message because you are subscribed to the Google
>>> >>> Groups "haskell-stack" group.
>>> >>> To unsubscribe from this group and stop receiving emails from it, send
>>> >>> an email to [email protected].
>>> >>> To post to this group, send email to [email protected].
>>> >>> To view this discussion on the web visit
>>> >>> https://groups.google.com/d/msgid/haskell-stack/5c46f622-a261-487b-96fd-3c9a86edb4ce%40googlegroups.com
>>> >>> <https://groups.google.com/d/msgid/haskell-stack/5c46f622-a261-487b-96fd-3c9a86edb4ce%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> >>> .
>>> >>> For more options, visit https://groups.google.com/d/optout.
>>> >>>
>>> >>
>>> >> --
>>> >> You received this message because you are subscribed to the Google Groups
>>> >> "haskell-stack" group.
>>> >> To unsubscribe from this group and stop receiving emails from it, send an
>>> >> email to [email protected].
>>> >> To post to this group, send email to [email protected].
>>> >> To view this discussion on the web visit
>>> >> https://groups.google.com/d/msgid/haskell-stack/CAKA2Jg%2BdzK7rBp9cJ6aLF1UbkBRkntFDWpnxa4KGbQ2VXCgVhQ%40mail.gmail.com
>>> >> <https://groups.google.com/d/msgid/haskell-stack/CAKA2Jg%2BdzK7rBp9cJ6aLF1UbkBRkntFDWpnxa4KGbQ2VXCgVhQ%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>> >> .
>>> >>
>>> >> For more options, visit https://groups.google.com/d/optout.
>>> >>
>>> >
>>> > --
>>> > You received this message because you are subscribed to the Google Groups
>>> > "haskell-stack" group.
>>> > To unsubscribe from this group and stop receiving emails from it, send an
>>> > email to [email protected].
>>> > To post to this group, send email to [email protected].
>>> > To view this discussion on the web visit
>>> > https://groups.google.com/d/msgid/haskell-stack/CAEYHaY52y5QuFu1QL3fTcgKE57F4hTzT5Au9Vc3Nr6gAstoixA%40mail.gmail.com
>>> > <https://groups.google.com/d/msgid/haskell-stack/CAEYHaY52y5QuFu1QL3fTcgKE57F4hTzT5Au9Vc3Nr6gAstoixA%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>> > .
>>> >
>>> > For more options, visit https://groups.google.com/d/optout.
>>> >
>>
>> --
>> You received this message because you are subscribed to the Google Groups 
>> "haskell-stack" group.
>> To unsubscribe from this group and stop receiving emails from it, send an 
>> email to [email protected].
>> To post to this group, send email to [email protected].
>> To view this discussion on the web visit 
>> https://groups.google.com/d/msgid/haskell-stack/1468257939-sup-6083%40sabre.
>> For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups 
"haskell-stack" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/haskell-stack/CAKcFpmKUjieorfDN_qx2ocNNW3xzF%3DfEOnkWvYT5z%2BCSrKsThg%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to