This patch adds a new script
hadrian/ghci which can be used to load GHC into GHCi. It supersedes the
There are also other small commits which were needed to fix dynamic linking and other aspects of hadrian. They should each be merged individually.
The patch works now but we need to cleanup the hacks. The following command loads the whole of GHC into GHCi in about 30 seconds on my very weedy machine.
ghci $(TERM=dumb ./hadrian/build.sh -j dump-args --build-root=_ghci -q) compiler/main/HscMain.hs
We have to build shared objects in case the user has a dynamically linked GHCi. However, the only way I could work out how to do this was to change the user setting to enable the dynamic way. This shouldn't be necessary but will probably need a lot of refactoring. Rules for
.dyn_ofiles should always be registered and selecting ways should just
needdifferent files rather than the current implicit way where they are all tied to the
-rpathhack that @DavidEichmann suggested looks quite dodgy, but why? It might be right.
Dynamic objects need to be built with the stage0 compiler. At the moment the name of the
.sois wrong if you enable this so I have hard coded the name into a few places.
Hacks like this need to be guarded by CPP or re-enabled if they work now.
-v0is passed to cabal should be configured by an env var - !509 (diffs)
Shake seems to output a control sequence to change the tab name even when
-qis passed which is why the
TERM=dumbenv var has to be set.