[GHC] #13604: ghci no longer loads dynamic .o files by default if they were built with -O

[GHC]

#13604: ghci no longer loads dynamic .o files by default if they were built with -O
-------------------------------------+-------------------------------------
        Reporter:  George            |                Owner:  dfeuer
            Type:  bug               |               Status:  patch
        Priority:  highest           |            Milestone:  8.2.2
       Component:  Compiler          |              Version:  8.2.1-rc1
      Resolution:                    |             Keywords:
                                     |  RecompilationCheck
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Runtime           |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):  Phab:D4123
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by elaforge):

 I don't totally understand the point about the debugger.  I thought ghci
 always loads binary if it can?  I usually use the * syntax for :load, so
 the current module is always interpreted, so I can see private names if
 there an export list.  Hasn't it always been true that to set a breakpoint
 you have to force the module to load as bytecode, either with * or by
 touching it so it thinks the binary is out of date?  I don't really use
 the debugger so I might be missing some detail.

 For context, I'm loading modules in two situations: one is from command
 line ghci, where I'm loading test modules, which were compiled with -fhpc.
 I also link the modules into a test binary, which I do I want -fhpc for so
 I can get coverage, but when testing from ghci I don't care about that
 stuff, I just want it to load the binary modules, not recompile everything
 as bytecode every time.

 The other situation is that I use the GHC API to load modules into a
 running application.  Those modules are compiled with -O, and I use
 GHC.setSessionDynFlags to get the same flags used to compile them when
 compiling the application itself so I can load them.  But the GHCI API
 then goes and filters out -O, making the flags different... if I'm
 remembering the results of my research correctly.  After that, I'll give
 it some expressions to evaluate, or maybe reload some modules as bytecode,
 just like you might do in ghci.  Similar to the -fhpc case, I don't
 actually care that the interpreted code is not optimized, I just want to
 load the binary modules.

 My suggestion was to turn off the thing that filters the flags.  Of course
 even if it retains -O it doesn't mean the bytecode interpreter can
 magically do optimizations, so it would be a bit of a lie.  But it seems
 like the lie is not so bad.  It would be optimizing if it could, and it
 will act as if the flag is set for the purposes of loading modules, but by
 its nature bytecode is not optimized, so it just doesn't apply when it
 compiles bytecode.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13604#comment:43>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler