[GHC] #15938: Hadrian's recompilation check is extremely slow

[GHC]

#15938: Hadrian's recompilation check is extremely slow
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:  alpmestan
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.8.1
       Component:  Build System      |              Version:  8.6.2
  (Hadrian)                          |
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by alpmestan):

 I'm looking into Andrey's suggestion (which I have been meaning to do for
 a while, ever since we implemented the "new" `Rules.Library` module). I
 did some profiling myself and the picture becomes even clearer if we hide
 shake from it:

 {{{
 COST CENTRE              MODULE                   SRC
 %time %alloc

 compilePackage           Rules.Compile
 src/Rules/Compile.hs:(13,1)-(36,77)        49.2   78.4
 split                    Data.List.Extra
 src/Data/List/Extra.hs:(569,1)-(571,48)    41.7    8.6
 modifyVar                Control.Concurrent.Extra
 src/Control/Concurrent/Extra.hs:191:1-36    2.3    1.1
 buildPackageDependencies Rules.Dependencies
 src/Rules/Dependencies.hs:(15,1)-(35,52)    1.5    2.9
 generatePackageCode      Rules.Generate
 src/Rules/Generate.hs:(102,1)-(150,49)      0.5    1.6
 configurePackage         Rules.Register
 src/Rules/Register.hs:(24,1)-(27,38)        0.4    1.5
 }}}

 While Shake itself might benefit from some optimisations for those paths
 manipulations that are involved here, clearly our number one problem is
 the enumeration of contexts for `compilePackage` that we do in `Rules.hs`,
 as you pointed out @snowleopard. I'm going to migrate `Rules.Compile` over
 to the enumeration-less, parsing-based style the best I can and report
 back.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15938#comment:27>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler