[GHC] #10844: CallStack should not be inlined

[GHC]

#10844: CallStack should not be inlined
-------------------------------------+-------------------------------------
        Reporter:  nomeata           |                   Owner:  gridaphobe
            Type:  task              |                  Status:  new
        Priority:  normal            |               Milestone:
       Component:  Compiler          |                 Version:  7.10.2
      Resolution:                    |                Keywords:
Operating System:  Unknown/Multiple  |            Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |               Test Case:
      Blocked By:                    |                Blocking:
 Related Tickets:                    |  Differential Revisions:
-------------------------------------+-------------------------------------

Comment (by gridaphobe):

 Just copying some notes from our discussion on phabricator for easy
 reference:

 gridaphobe:
 > - The increase in binary sizes (at least for the atom benchmark) is
 coming from a handful of modules in ghc-prim and base
 > - GHC/Types.o has increased by about 5K. It's now the home of the SrcLoc
 and CallStack types, which together define 7-8 record selectors, so I
 think this may not be too surprising.
 > - Control/Applicative.o, Data/Either.o, Data/Foldable.o, and a few
 others in base have increased in by 2-5K each. Every one of these modules
 contains at least one CallStack in the Core.
 > - The other modules I've looked at that have not increased in size do
 not have a CallStack, so it seems pretty clear that the increase *is* due
 to the CallStack.
 > - Finally, I think you're correct that the CallStacks are being inlined
 sometimes, because I found a CallStack referencing GHC.Base in
 Data.Foldable.

 nomeata:
 > I’m not quite sure what the best way is to prevent the sharing. Maybe
 the type checker, which produces the CallStack values, can be told to add
 the (core equivalent) of a NOINLINE pragma to them?


 ---

 I agree that CallStacks should not be inlined. Is there a core equivalent
 of a NOINLINE pragma? If so, it ought the be straightforward to have the
 desugarer attach it.

 Also, is there any extra cost to lifting all of the int and string
 components of a `SrcLoc` to top-level binders (which seems similarly
 pointless)? Or does it just clutter the Core output?

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10844#comment:2>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler