GHCi warns about -eventlog even though it's sometimes necessary

changed weight to 5

@akio I can't reproduce your bug.

Could you please supply some example code where running ghci with -eventlog does make a difference.

Here's what I've tried:

$ cat Test.hs
main = return ()

$ ghc-7.6.3 -fforce-recomp -eventlog -c Test.hs

$ ghc-7.6.3 --interactive Test.o
GHCi, version 7.6.3: http://www.haskell.org/ghc/  :? for help
Loading package ghc-prim ... linking ... done.
Loading package integer-gmp ... linking ... done.
Loading package base ... linking ... done.
Loading object (static) Test.o ... done
final link ... done
Leaving GHCi.

Seems to work fine.

Here is a test session (with GHC 7.10.1):

% cat Test.hs
main = return ()
% ghc -c -eventlog Test.hs
% ghci Test.hs
GHCi, version 7.10.1.1: http://www.haskell.org/ghc/  :? for help
[1 of 1] Compiling Main             ( Test.hs, interpreted ) [flags changed]
Ok, modules loaded: Main.
*Main> 
Leaving GHCi.
% ghci Test.hs -eventlog
GHCi, version 7.10.1.1: http://www.haskell.org/ghc/  :? for help
Warning: -debug, -threaded and -ticky are ignored by GHCi
Ok, modules loaded: Main.
Prelude Main> 
Leaving GHCi.

So GHCi decides to re-use the object file only when it's given -eventlog. Perhaps the recompilation checker should ignore this flag, if the flag doesn't actually affect how a module is compiled?

akio, you seem to have something special about your environment (DYNAMIC_BY_DEFAULT?) I had to add -dynamic to the ghc invocations to reproduce.

But a simpler test is

rwbarton@morphism:/tmp$ ghc -c Test.hs 
rwbarton@morphism:/tmp$ ghc -c Test.hs -threaded
compilation IS NOT required
rwbarton@morphism:/tmp$ ghc -c Test.hs -debug
compilation IS NOT required
rwbarton@morphism:/tmp$ ghc -c Test.hs -eventlog
rwbarton@morphism:/tmp$

Technically the eventlog way is not actually RTS-only because it adds -DTRACING to the C compiler and C preprocessor options. If you had CPP in Test.hs that checked whether TRACING was defined, then recompiling the module to an object file (either in ghci or by ghc) really would have been necessary.

So, there are a few possible resolutions:

Make WayEventLog a non-RTS only way in wayRTSOnly, which would suppress this ghci warning, on the grounds that modules can detect (via CPP) whether they were built with -eventlog. I think this would prevent linking -eventlog object files with non--eventlog ones, though.
Don't change wayRTSOnly, and just change the test for that ghci warning to check something other than wayRTSOnly (e.g., treat WayEventLog specially). However this leaves us in the current mildly inconsistent state of affairs that -eventlog is considered in some sense (and is documented to be) a link-time only option, when in fact it affects compilation as well.
Implement WayEventLog in the same manner as WayThreaded and WayDebug. I don't know exactly what is (perhaps thomie can comment) but the latter two ways are not built by automatically enabling CPP options in ghc, even though the RTS is in fact full of #ifdef THREADED and #ifdef DEBUG. I assume there is some special logic in the build system for these. This potentially breaks people who were using #ifdef TRACING in their Haskell source, but hopefully there aren't any (it is undocumented after all).

Personally I lean towards 3 if it isn't technically difficult to implement.

Replying to [ticket:8123#comment:102448 rwbarton]:

akio, you seem to have something special about your environment (DYNAMIC_BY_DEFAULT?) I had to add -dynamic to the ghc invocations to reproduce.

Oops, yes, I was using a ghc with ("GHC Dynamic","NO").

added error messages label

added eventlog label

added Pnormal label

Trac field	Value
Version	7.6.3
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	GHCi
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

GHCi warns about -eventlog even though it's sometimes necessary

Child items 0

Activity