restored support for incremental compilation of scripts to C++

src/script_opt/CPP/Compile.h

@@ -585,10 +585,6 @@ private:
 	// function name, and maps to the C++ name.
 	std::unordered_map<std::string, std::string> compiled_simple_funcs;
 
-	// Maps those to their associated files - used to make add-C++ body
-	// hashes distinct.
-	std::unordered_map<std::string, std::string> cf_locs;
-
 	// Maps function bodies to the names we use for them.
 	std::unordered_map<const Stmt*, std::string> body_names;
 

src/script_opt/CPP/DeclFunc.cc

@@ -99,9 +99,6 @@ void CPPCompile::CreateFunction(const FuncTypePtr& ft, const ProfileFunc* pf, co
 		// for lambdas that don't take any arguments, but that
 		// seems potentially more confusing than beneficial.
 		compiled_funcs.emplace(fname);
-
-		auto loc_f = script_specific_filename(body);
-		cf_locs[fname] = loc_f;
 		}
 
 	auto h = pf->HashVal();

src/script_opt/CPP/Driver.cc

@@ -197,10 +197,10 @@ void CPPCompile::Compile(bool report_uncompilable)
 
 void CPPCompile::GenProlog()
 	{
-	if ( addl_tag == 0 )
+	if ( addl_tag <= 1 )
-		{
+		// This is either a compilation via gen-C++, or
+		// one using add-C++ and an empty CPP-gen.cc file.
 		Emit("#include \"zeek/script_opt/CPP/Runtime.h\"\n");
-		}
 
 	Emit("namespace zeek::detail { //\n");
 	Emit("namespace CPP_%s { // %s\n", Fmt(addl_tag), working_dir);
@@ -292,13 +292,6 @@ void CPPCompile::RegisterCompiledBody(const string& f)
 
 	events = string("{") + events + "}";
 
-	if ( addl_tag > 0 )
-		// Hash in the location associated with this compilation
-		// pass, to get a final hash that avoids conflicts with
-		// identical-but-in-a-different-context function bodies
-		// when compiling potentially conflicting additional code.
-		h = merge_p_hashes(h, p_hash(cf_locs[f]));
-
 	auto fi = func_index.find(f);
 	ASSERT(fi != func_index.end());
 	auto type_signature = casting_index[fi->second];

src/script_opt/CPP/Func.h

@@ -105,6 +105,7 @@ struct CompiledScript
 	CPPStmtPtr body;
 	int priority;
 	std::vector<std::string> events;
+	void (*finish_init_func)();
 	};
 
 // Maps hashes to compiled information.

src/script_opt/CPP/InitsInfo.cc

@@ -239,14 +239,14 @@ void FuncConstInfo::InitializerVals(std::vector<std::string>& ivs) const
 	{
 	auto f = fv->AsFunc();
 	const auto& fn = f->Name();
+	const auto& bodies = f->GetBodies();
 
 	ivs.emplace_back(Fmt(type));
 	ivs.emplace_back(Fmt(c->TrackString(fn)));
+	ivs.emplace_back(to_string(bodies.size()));
 
 	if ( ! c->NotFullyCompilable(fn) )
 		{
-		const auto& bodies = f->GetBodies();
-
 		for ( const auto& b : bodies )
 			{
 			auto h = c->BodyHash(b.stmts.get());

src/script_opt/CPP/RuntimeInitSupport.cc

@@ -62,9 +62,10 @@ void register_type__CPP(TypePtr t, const string& name)
 	id->MakeType();
 	}
 
-void register_body__CPP(CPPStmtPtr body, int priority, p_hash_type hash, vector<string> events)
+void register_body__CPP(CPPStmtPtr body, int priority, p_hash_type hash, vector<string> events,
+                        void (*finish_init)())
 	{
-	compiled_scripts[hash] = {move(body), priority, move(events)};
+	compiled_scripts[hash] = {move(body), priority, move(events), finish_init};
 	}
 
 void register_lambda__CPP(CPPStmtPtr body, p_hash_type hash, const char* name, TypePtr t,
@@ -89,7 +90,7 @@ void register_lambda__CPP(CPPStmtPtr body, p_hash_type hash, const char* name, T
 	if ( ! has_captures )
 		// Note, no support for lambdas that themselves refer
 		// to events.
-		register_body__CPP(body, 0, hash, {});
+		register_body__CPP(body, 0, hash, {}, nullptr);
 	}
 
 void register_scripts__CPP(p_hash_type h, void (*callback)())
@@ -190,11 +191,12 @@ Func* lookup_bif__CPP(const char* bif)
 	return b ? b->GetVal()->AsFunc() : nullptr;
 	}
 
-FuncValPtr lookup_func__CPP(string name, vector<p_hash_type> hashes, const TypePtr& t)
+FuncValPtr lookup_func__CPP(string name, int num_bodies, vector<p_hash_type> hashes,
+                            const TypePtr& t)
 	{
 	auto ft = cast_intrusive<FuncType>(t);
 
-	if ( hashes.empty() )
+	if ( static_cast<int>(hashes.size()) < num_bodies )
 		{
 		// This happens for functions that have at least one
 		// uncompilable body.

src/script_opt/CPP/RuntimeInitSupport.h

@@ -37,7 +37,7 @@ extern void register_type__CPP(TypePtr t, const std::string& name);
 // relevant for the function body, which should be registered if the
 // function body is going to be used.
 extern void register_body__CPP(CPPStmtPtr body, int priority, p_hash_type hash,
-                               std::vector<std::string> events);
+                               std::vector<std::string> events, void (*finish_init)());
 
 // Registers a lambda body as associated with the given hash.  Includes
 // the name of the lambda (so it can be made available as a quasi-global
@@ -67,7 +67,8 @@ extern Func* lookup_bif__CPP(const char* bif);
 // returns an associated FuncVal.  It's a fatal error for the hash
 // not to exist, because this function should only be called by compiled
 // code that has ensured its existence.
-extern FuncValPtr lookup_func__CPP(std::string name, std::vector<p_hash_type> h, const TypePtr& t);
+extern FuncValPtr lookup_func__CPP(std::string name, int num_bodies, std::vector<p_hash_type> h,
+                                   const TypePtr& t);
 
 // Returns the record corresponding to the given name, as long as the
 // name is indeed a record type.  Otherwise (or if the name is nil)

src/script_opt/CPP/RuntimeInits.cc

@@ -147,13 +147,14 @@ void CPP_IndexedInits<T>::Generate(InitsManager* im, std::vector<FuncValPtr>& iv
 	auto iv_end = init_vals.end();
 	auto t = *(iv_it++);
 	auto fn = im->Strings(*(iv_it++));
+	auto num_bodies = *(iv_it++);
 
 	std::vector<p_hash_type> hashes;
 
 	while ( iv_it != iv_end )
 		hashes.push_back(im->Hashes(*(iv_it++)));
 
-	ivec[offset] = lookup_func__CPP(fn, hashes, im->Types(t));
+	ivec[offset] = lookup_func__CPP(fn, num_bodies, hashes, im->Types(t));
 	}
 
 template <class T>

src/script_opt/ProfileFunc.cc

@@ -23,7 +23,13 @@ p_hash_type p_hash(const Obj* o)
 	return p_hash(d.Description());
 	}
 
-std::string script_specific_filename(const StmtPtr& body)
+// Returns a filename associated with the given function body.  Used to
+// provide distinctness to identical function bodies seen in separate,
+// potentially conflicting incremental compilations.  An example of this
+// is a function named foo() that calls bar(), for which in two different
+// compilation contexts bar() has differing semantics, even though foo()'s
+// (shallow) semantics are the same.
+static std::string script_specific_filename(const Stmt* body)
 	{
 	// The specific filename is taken from the location filename, making
 	// it absolute if necessary.
@@ -52,12 +58,15 @@ std::string script_specific_filename(const StmtPtr& body)
 	return bl_f;
 	}
 
-p_hash_type script_specific_hash(const StmtPtr& body, p_hash_type generic_hash)
+// Returns a incremental-compilation-specific hash for the given function
+// body, given its non-specific hash is "generic_hash".
+static p_hash_type script_specific_hash(const Stmt* body, p_hash_type generic_hash)
 	{
 	auto bl_f = script_specific_filename(body);
 	return merge_p_hashes(generic_hash, p_hash(bl_f));
 	}
 
+
 ProfileFunc::ProfileFunc(const Func* func, const StmtPtr& body, bool _abs_rec_fields)
 	{
 	profiled_func = func;
@@ -698,6 +707,9 @@ void ProfileFuncs::ComputeProfileHash(std::shared_ptr<ProfileFunc> pf)
 	for ( auto i : pf->AdditionalHashes() )
 		h = merge_p_hashes(h, i);
 
+	if ( ! pf->Stmts().empty() )
+		h = script_specific_hash(pf->Stmts()[0], h);
+
 	pf->SetHashVal(h);
 	}
 

src/script_opt/ProfileFunc.h

@@ -74,17 +74,6 @@ inline p_hash_type merge_p_hashes(p_hash_type h1, p_hash_type h2)
 	return h1 ^ (h2 + 0x9e3779b9 + (h1 << 6) + (h1 >> 2));
 	}
 
-// Returns a filename associated with the given function body.  Used to
-// provide distinctness to identical function bodies seen in separate,
-// potentially conflicting incremental compilations.  This is only germane
-// for allowing incremental compilation of subsets of the test suite, so
-// if we decide to forgo that capability, we can remove this.
-extern std::string script_specific_filename(const StmtPtr& body);
-
-// Returns a incremental-compilation-specific hash for the given function
-// body, given it's non-specific hash is "generic_hash".
-extern p_hash_type script_specific_hash(const StmtPtr& body, p_hash_type generic_hash);
-
 // Class for profiling the components of a single function (or expression).
 class ProfileFunc : public TraversalCallback
 	{

src/script_opt/ScriptOpt.cc

@@ -350,17 +350,8 @@ static void report_CPP()
 		auto name = f.Func()->Name();
 		auto hash = f.Profile()->HashVal();
 		bool have = compiled_scripts.count(hash) > 0;
-		auto specific = "";
 
-		if ( ! have )
+		printf("script function %s (hash %llu): %s\n", name, hash, have ? "yes" : "no");
-			{
-			hash = script_specific_hash(f.Body(), hash);
-			have = compiled_scripts.count(hash) > 0;
-			if ( have )
-				specific = " - specific";
-			}
-
-		printf("script function %s (hash %llu%s): %s\n", name, hash, specific, have ? "yes" : "no");
 
 		if ( have )
 			already_reported.insert(hash);
@@ -387,12 +378,6 @@ static void use_CPP()
 		auto hash = f.Profile()->HashVal();
 		auto s = compiled_scripts.find(hash);
 
-		if ( s == compiled_scripts.end() )
-			{ // Look for script-specific body.
-			hash = script_specific_hash(f.Body(), hash);
-			s = compiled_scripts.find(hash);
-			}
-
 		if ( s != compiled_scripts.end() )
 			{
 			auto b = s->second.body;
@@ -418,6 +403,10 @@ static void use_CPP()
 				auto h = event_registry->Register(e);
 				h->SetUsed();
 				}
+
+			auto finish = s->second.finish_init_func;
+			if ( finish )
+				(*finish)();
 			}
 		}