Fix ZST handling in ccall.cpp

Author: JSmith2023

src/ccall.cpp

@@ -919,7 +919,8 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
     jl_svec_t *tt = ((jl_datatype_t *)at)->parameters;
     size_t nargt = jl_svec_len(tt);
     SmallVector<llvm::Type*, 0> argtypes;
-    SmallVector<Value *, 8> argvals(nargt);
+    SmallVector<Value *, 8> argvals; //do not presize argvals, we will push_back only non-void arguments
+    argvals.reserve(nargt);
     for (size_t i = 0; i < nargt; ++i) {
         jl_value_t *tti = jl_svecref(tt,i);
         bool toboxed;
@@ -933,9 +934,16 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
         jl_value_t *argi = args[4 + i];
         jl_cgval_t arg = emit_expr(ctx, argi);
 
-        Value *v = julia_to_native(ctx, t, toboxed, tti, NULL, arg, false, i);
-        bool issigned = jl_signed_type && jl_subtype(tti, (jl_value_t*)jl_signed_type);
-        argvals[i] = llvm_type_rewrite(ctx, v, t, issigned);
+        Value *v = julia_to_native(ctx, t, toboxed, tti, NULL, arg, false, i); //1. If Void -> SKIP (do not push to argvals) 2. If empty struct -> push UndefValue 3. Normal case -> push value
+        if (t->isVoidTy()) {
+            continue;
+        }
+        if (v == NULL){
+            argvals.push_back(UndefValue::get(t));
+        } else{
+            bool issigned = jl_signed_type && jl_subtype(tti, (jl_value_t*)jl_signed_type);
+            argvals.push_back(llvm_type_rewrite(ctx, v, t, issigned));
+        }
     }
 
     // Determine return type
@@ -963,11 +971,11 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
         // stringify arguments
         std::string arguments;
         raw_string_ostream argstream(arguments);
+        int printed_args = 0;
         for (SmallVector<Type *, 0>::iterator it = argtypes.begin(); it != argtypes.end(); ++it) {
-            if (it != argtypes.begin())
-                argstream << ",";
+            if ((*it)-> isVoidTy()) continue;
+            if (printed_args++ > 0) argstream << ", ";
             (*it)->print(argstream);
-            argstream << " ";
         }
 
         // stringify return type
@@ -991,17 +999,13 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
         if (!Mod) {
             std::string compat_arguments;
             raw_string_ostream compat_argstream(compat_arguments);
+            int printed_args = 0;
             for (size_t i = 0; i < nargt; ++i) {
-                if (i > 0)
-                    compat_argstream << ",";
                 jl_value_t *tti = jl_svecref(tt, i);
-                Type *t;
-                if (jl_is_cpointer_type(tti))
-                    t = ctx.types().T_size;
-                else
-                    t = argtypes[i];
+                Type *t = (jl_is_cpointer_type(tti)) ? ctx.types().T_size : argtypes[i];
+                if (t->isVoidTy()) continue;
+                if (printed_args++ > 0) compat_argstream << ", ";
                 t->print(compat_argstream);
-                compat_argstream << " ";
             }
 
             std::string compat_rstring;
@@ -1084,13 +1088,18 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
 
     // backwards compatibility: support for IR with integer pointers
     bool mismatched_pointers = false;
+    int arg_idx = 0;
     for (size_t i = 0; i < nargt; ++i) {
         jl_value_t *tti = jl_svecref(tt, i);
+        Type *t = argtypes[i];
+        if (t->isVoidTy())
+            continue;
         if (jl_is_cpointer_type(tti) &&
-            !f->getFunctionType()->getParamType(i)->isPointerTy()) {
+            !f->getFunctionType()->getParamType(arg_idx)->isPointerTy()) {
             mismatched_pointers = true;
             break;
         }
+        arg_idx++;
     }
     if (mismatched_pointers) {
         if (jl_options.depwarn) {
@@ -1128,9 +1137,13 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
         BasicBlock *entry = BasicBlock::Create(ctx.builder.getContext(), "", wrapper);
         IRBuilder<> irbuilder(entry);
         SmallVector<Value *, 0> wrapper_args;
+        arg_idx = 0;
         for (size_t i = 0; i < nargt; ++i) {
             jl_value_t *tti = jl_svecref(tt, i);
-            Value *v = wrapper->getArg(i);
+            Type *t = argtypes[i];
+            if (t->isVoidTy())
+                continue;
+            Value *v = wrapper->getArg(arg_idx++);
             if (jl_is_cpointer_type(tti))
                 v = irbuilder.CreatePtrToInt(v, ctx.types().T_size);
             wrapper_args.push_back(v);
@@ -1150,18 +1163,21 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
 
     // verify the function type
     assert(f->getReturnType() == rettype);
-    int i = 0;
+    arg_idx = 0;
     for (SmallVector<Type *, 0>::iterator it = argtypes.begin(); it != argtypes.end();
-         ++it, ++i) {
-        if (*it != f->getFunctionType()->getParamType(i)) {
+         ++it) {
+        if ((*it)->isVoidTy())
+            continue;
+        if (*it != f->getFunctionType()->getParamType(arg_idx)) {
             std::string message;
             raw_string_ostream stream(message);
-            stream << "Malformed llvmcall: argument " << i + 1 << " type "
-                   << *f->getFunctionType()->getParamType(i)
+            stream << "Malformed llvmcall: argument " << arg_idx + 1 << " type "
+                   << *f->getFunctionType()->getParamType(arg_idx)
                    << " does not match expected argument type " << **it;
             emit_error(ctx, stream.str());
             return jl_cgval_t();
         }
+        arg_idx++;
     }
 
     // copy module properties that should always match
@@ -1183,7 +1199,12 @@ static jl_cgval_t emit_llvmcall(jl_codectx_t &ctx, jl_value_t **args, size_t nar
     def->setLinkage(GlobalVariable::LinkOnceODRLinkage);
 
     // generate a call
-    FunctionType *decl_typ = FunctionType::get(rettype, argtypes, def->isVarArg());
+    SmallVector<Type *, 0> filtered_argtypes;
+    for (auto t : argtypes) {
+        if (!t->isVoidTy())
+            filtered_argtypes.push_back(t);
+    }
+    FunctionType *decl_typ = FunctionType::get(rettype, filtered_argtypes, def->isVarArg());
     Function *decl = Function::Create(decl_typ, def->getLinkage(), def->getAddressSpace(),
                                       def->getName(), jl_Module);
     decl->setAttributes(def->getAttributes());

test/llvmcall.jl

@@ -218,3 +218,16 @@ s = MyStruct()
 @test eltype(supertype(Core.LLVMPtr{UInt8,1})) <: UInt8
 @test s.kern == 0
 @test reinterpret(Int, s.ptr) == 0
+
+# Issue #60063: llvmcall segfault/parser error with zero-sized argument (Nothing)
+module LLVMCallTest
+    using Test
+    let
+        f(x::T) where T = Base.llvmcall("ret i8 %0", Int8, Tuple{T}, x)
+        err = try f(nothing) catch e e end
+        @test err isa ErrorException
+        #verify the error is about the LLVM assembly failure and references the undefined symbol
+        @test occursin("Failed to parse LLVM assembly", err.msg)
+        @test occursin("%0", err.msg)
+        end
+end #module LLVMCallTest