Re: [GHC] #12620: Allow the user to prevent floating and CSE

29 Sep 2016

      #12620: Allow the user to prevent floating and CSE
-------------------------------------+-------------------------------------
        Reporter:  nomeata           |                Owner:
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.0.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #9520, #8457      |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by edsko):

 Replying to [comment:9 tomjaguarpaw]:
...
I'm very glad to see full laziness getting some attention (...) I have
 even asked whether it is an optimization worth performing at all, though I
 conclude that it is:
* https://stackoverflow.com/questions/35115172/why-is-full-laziness-a
 -default-optimization/35115664
...
However I do not think this suggestion is the right approach.  (...)
 The proposed `nofloat` keyword is just adding additional complexity over a
Yup, I cite this in the blog post :)

 transformation which itself is introducing too much complexity.  I'm very
 concerned about the idea.

 I agree that it would be preferable not to "program the optimizer" when
 writing Haskell code. That's another reason in fact why I prefer
 `noupdate` over `nofloat`, beacuse actually `noupdate` goes beyond full
 laziness. Consider this example from the blog post:

 {{{#!hs
 retry :: IO a -> IO a
 retry io = catch io (\(_ :: SomeException) -> retry io)

 main :: IO ()
 main = retry $ ni_mapM_ print [1..1000000]
 }}}

 This program has a memory leak, but it's nothing to do with full laziness
 here. Now admittedly we could turn this into a full laziness issue by
 giving the argument to `retry` a dummy unit argument or something like
 that, so that we write

 {{{#!hs
 retry :: (() -> IO a) -> IO a
 retry io = catch (io ()) (\(_ :: SomeException) -> retry io)

 main :: IO ()
 main = retry $ \() -> ni_mapM_ print [1..1000000]
 }}}

 or something like that, but then you would have to do that in every single
 function that duplicates IO actions (think `forever`, `replicateM_`, etc.)
 Instead, we could mark that list as `noupdate` and the memory leak would
 be gone.

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/12620#comment:11
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler