Interface Files with Core Definitions

This commit adds three new flags

* -fwrite-if-simplified-core: Writes the whole core program into an interface
  file
* -fbyte-code-and-object-code: Generate both byte code and object code
  when compiling a file
* -fprefer-byte-code: Prefer to use byte-code if it's available when
  running TH splices.

The goal for including the core bindings in an interface file is to be able to restart the compiler pipeline
at the point just after simplification and before code generation. Once compilation is
restarted then code can be created for the byte code backend.
This can significantly speed up
start-times for projects in GHCi. HLS already implements its own version of these extended interface
files for this reason.

Preferring to use byte-code means that we can avoid some potentially
expensive code generation steps (see #21700)

* Producing object code is much slower than producing bytecode, and normally you
  need to compile with `-dynamic-too` to produce code in the static and dynamic way, the
  dynamic way just for Template Haskell execution when using a dynamically linked compiler.

* Linking many large object files, which happens once per splice, can be quite
  expensive compared to linking bytecode.

And you can get GHC to compile the necessary byte code so
`-fprefer-byte-code` has access to it by using
`-fbyte-code-and-object-code`.

Fixes #21067

4 files changed, 83 insertions, 9 deletions

diff --git a/compiler/GHC/Unit/Module/Graph.hs b/compiler/GHC/Unit/Module/Graph.hs
index f8b0bcc2c3..fcd6a63a28 100644
--- a/compiler/GHC/Unit/Module/Graph.hs
+++ b/compiler/GHC/Unit/Module/Graph.hs
@@ -282,12 +282,12 @@ showModMsg dflags recomp (ModuleNode _ mod_summary) =
     mod_str  = showPpr dflags mod ++ hscSourceString (ms_hsc_src mod_summary)
     dyn_file = op $ msDynObjFilePath mod_summary
     obj_file = op $ msObjFilePath mod_summary
+    files    = [ obj_file ]
+               ++ [ dyn_file | gopt Opt_BuildDynamicToo dflags ]
+               ++ [ "interpreted" | gopt Opt_ByteCodeAndObjectCode dflags ]
     message = case backendSpecialModuleSource (backend dflags) recomp of
                 Just special -> text special
-                Nothing ->
-                  if gopt Opt_BuildDynamicToo  dflags
-                    then text obj_file <> comma <+> text dyn_file
-                    else text obj_file
+                Nothing -> foldr1 (\ofile rest -> ofile <> comma <+> rest) (map text files)
 
 
 
diff --git a/compiler/GHC/Unit/Module/ModIface.hs b/compiler/GHC/Unit/Module/ModIface.hs
index 76cfff2b9f..1d5280f4fa 100644
--- a/compiler/GHC/Unit/Module/ModIface.hs
+++ b/compiler/GHC/Unit/Module/ModIface.hs
@@ -200,6 +200,11 @@ data ModIface_ (phase :: ModIfacePhase)
                 -- Ditto data constructors, class operations, except that
                 -- the hash of the parent class/tycon changes
 
+        mi_extra_decls :: Maybe [IfaceBindingX IfaceMaybeRhs IfaceTopBndrInfo],
+                -- ^ Extra variable definitions which are **NOT** exposed but when
+                -- combined with mi_decls allows us to restart code generation.
+                -- See Note [Interface Files with Core Definitions] and Note [Interface File with Core: Sharing RHSs]
+
         mi_globals  :: !(Maybe GlobalRdrEnv),
                 -- ^ Binds all the things defined at the top level in
                 -- the /original source/ code for this module. which
@@ -349,6 +354,7 @@ instance Binary ModIface where
                  mi_warns     = warns,
                  mi_anns      = anns,
                  mi_decls     = decls,
+                 mi_extra_decls = extra_decls,
                  mi_insts     = insts,
                  mi_fam_insts = fam_insts,
                  mi_rules     = rules,
@@ -392,6 +398,7 @@ instance Binary ModIface where
         lazyPut bh warns
         lazyPut bh anns
         put_ bh decls
+        put_ bh extra_decls
         put_ bh insts
         put_ bh fam_insts
         lazyPut bh rules
@@ -423,6 +430,7 @@ instance Binary ModIface where
         warns       <- {-# SCC "bin_warns" #-} lazyGet bh
         anns        <- {-# SCC "bin_anns" #-} lazyGet bh
         decls       <- {-# SCC "bin_tycldecls" #-} get bh
+        extra_decls <- get bh
         insts       <- {-# SCC "bin_insts" #-} get bh
         fam_insts   <- {-# SCC "bin_fam_insts" #-} get bh
         rules       <- {-# SCC "bin_rules" #-} lazyGet bh
@@ -446,6 +454,7 @@ instance Binary ModIface where
                  mi_fixities    = fixities,
                  mi_warns       = warns,
                  mi_decls       = decls,
+                 mi_extra_decls = extra_decls,
                  mi_globals     = Nothing,
                  mi_insts       = insts,
                  mi_fam_insts   = fam_insts,
@@ -494,6 +503,7 @@ emptyPartialModIface mod
                mi_fam_insts   = [],
                mi_rules       = [],
                mi_decls       = [],
+               mi_extra_decls = Nothing,
                mi_globals     = Nothing,
                mi_hpc         = False,
                mi_trust       = noIfaceTrustInfo,
@@ -541,12 +551,13 @@ emptyIfaceHashCache _occ = Nothing
 -- avoid major space leaks.
 instance (NFData (IfaceBackendExts (phase :: ModIfacePhase)), NFData (IfaceDeclExts (phase :: ModIfacePhase))) => NFData (ModIface_ phase) where
   rnf (ModIface f1 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11 f12
-                f13 f14 f15 f16 f17 f18 f19 f20 f21 f22 f23) =
+                f13 f14 f15 f16 f17 f18 f19 f20 f21 f22 f23 f24) =
     rnf f1 `seq` rnf f2 `seq` f3 `seq` f4 `seq` f5 `seq` f6 `seq` rnf f7 `seq` f8 `seq`
-    f9 `seq` rnf f10 `seq` rnf f11 `seq` f12 `seq` rnf f13 `seq` rnf f14 `seq` rnf f15 `seq`
-    rnf f16 `seq` f17 `seq` rnf f18 `seq` rnf f19 `seq` f20 `seq` f21 `seq` f22 `seq` rnf f23
+    f9 `seq` rnf f10 `seq` rnf f11 `seq` rnf f12 `seq` f13 `seq` rnf f14 `seq` rnf f15 `seq` rnf f16 `seq`
+    rnf f17 `seq` f18 `seq` rnf f19 `seq` rnf f20 `seq` f21 `seq` f22 `seq` f23 `seq` rnf f24
     `seq` ()
 
+
 instance NFData (ModIfaceBackend) where
   rnf (ModIfaceBackend f1 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11 f12 f13)
     = rnf f1 `seq` rnf f2 `seq` rnf f3 `seq` rnf f4 `seq`
diff --git a/compiler/GHC/Unit/Module/Status.hs b/compiler/GHC/Unit/Module/Status.hs
index 6f926e3fb2..49851b74eb 100644
--- a/compiler/GHC/Unit/Module/Status.hs
+++ b/compiler/GHC/Unit/Module/Status.hs
@@ -10,13 +10,13 @@ import GHC.Unit.Module.ModGuts
 import GHC.Unit.Module.ModIface
 
 import GHC.Utils.Fingerprint
-import GHC.Linker.Types
 import GHC.Utils.Outputable
+import GHC.Unit.Home.ModInfo
 
 -- | Status of a module in incremental compilation
 data HscRecompStatus
     -- | Nothing to do because code already exists.
-    = HscUpToDate ModIface (Maybe Linkable)
+    = HscUpToDate ModIface HomeModLinkable
     -- | Recompilation of module, or update of interface is required. Optionally
     -- pass the old interface hash to avoid updating the existing interface when
     -- it has not changed.
diff --git a/compiler/GHC/Unit/Module/WholeCoreBindings.hs b/compiler/GHC/Unit/Module/WholeCoreBindings.hs
new file mode 100644
index 0000000000..8e84abbf57
--- /dev/null
+++ b/compiler/GHC/Unit/Module/WholeCoreBindings.hs
@@ -0,0 +1,63 @@
+module GHC.Unit.Module.WholeCoreBindings where
+
+import GHC.Unit.Types (Module)
+import GHC.Unit.Module.Location
+import GHC.Iface.Syntax
+
+{-
+Note [Interface Files with Core Definitions]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+A interface file can optionally contain the definitions of all core bindings, this
+is enabled by the flag `-fwrite-if-simplified-core`.
+This provides everything needed in addition to the normal ModIface and ModDetails
+to restart compilation after typechecking to generate bytecode. The `fi_bindings` field
+is stored in the normal interface file and the other fields populated whilst loading
+the interface file.
+
+The lifecycle of a WholeCoreBindings typically proceeds as follows:
+
+1. The ModIface which contains mi_extra_decls is loaded from disk. A linkable is
+   created (which is headed by the `CoreBindings` constructor). This is an unhydrated set of bindings which
+   is currently unsuitable for linking, but at the point it is loaded, the ModIface
+   hasn't been hydrated yet (See Note [Hydrating Modules]) either so the CoreBindings constructor allows the delaying of converting
+   the WholeCoreBindings into a proper Linkable (if we ever do that). The CoreBindings constructor also
+   allows us to convert the WholeCoreBindings into multiple different linkables if we so desired.
+
+2. `initWholeCoreBindings` turns a WholeCoreBindings into a proper BCO linkable. This step combines together
+   all the necessary information from a ModIface, ModDetails and WholeCoreBindings in order to
+   create the linkable. The linkable created is a "LoadedBCOs" linkable, which
+   was introduced just for initWholeCoreBindings, so that the bytecode can be generated lazilly.
+   Using the `BCOs` constructor directly here leads to the bytecode being forced
+   too eagerly.
+
+3. Then when bytecode is needed, the LoadedBCOs value is inspected and unpacked and
+   the linkable is used as before.
+
+The flag `-fwrite-if-simplfied-core` determines whether the extra information is written
+to an interface file. The program which is written is the core bindings of the module
+after whatever simplification the user requested has been performed. So the simplified core bindings
+of the interface file agree with the optimisation level as reported by the interface
+file.
+
+Note [Size of Interface Files with Core Definitions]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+How much overhead does `-fwrite-if-simplfied-core` add to a typical interface file?
+As an experiment I compiled the `Cabal` library and `ghc` library (Aug 22) with
+
+| Project | .hi  | .hi (fat) | .o   |
+| --------| ---- | --------- | --   |
+| ghc     | 32M  | 68M       | 127M |
+| Cabal   | 3.2M | 9.8M      | 14M  |
+
+So the interface files gained in size but the end result was still smaller than
+the object files.
+
+-}
+
+data WholeCoreBindings = WholeCoreBindings
+            { wcb_bindings :: [IfaceBindingX IfaceMaybeRhs IfaceTopBndrInfo]
+            , wcb_module   :: Module
+            , wcb_mod_location :: ModLocation
+            }