Compilation failure on code analysis

[jcrapuchettes]

While attempting compilation of my company's code library tests, I ran into a compiler error. I have a working checkout at commit 22749e4.

Minimized code to produce the error:

struct Cube(StorageT)
{
    StorageT datastore;
}

auto test(Cube)(Cube cube)
{
    struct TestStruct(StorageType)
    {
        struct InnerStruct {}
        InnerStruct[] dimErrors;    //InnerStruct is the cause. Changing to some other type will allow compilation to work.
    }
    return TestStruct!(typeof(cube.datastore))();
}

void main()
{
    auto dataCube = Cube!double();
    test(dataCube);
}

Compiler output:

~/ldc/build/bin/ldmd2 -v ldc_test.d 
config    /home/jonathan/ldc/build/bin/ldc2.conf
parse     ldc_test
importall ldc_test
import    object    (/home/jonathan/ldc/build/import/object.di)
semantic  ldc_test
entry     main          ldc_test.d
semantic2 ldc_test
semantic3 ldc_test
code      ldc_test
0  ldc2            0x00000000012b0a6f
1  ldc2            0x00000000012b0ed9
2  libpthread.so.0 0x00007f24884ebbd0
3  ldc2            0x00000000006d1fe2 TypeInfoStructDeclaration::llvmDefine() + 226
4  ldc2            0x00000000006d14d4 DtoDeclareTypeInfo(TypeInfoDeclaration*) + 244
5  ldc2            0x00000000006d16fe DtoResolveTypeInfo(TypeInfoDeclaration*) + 462
6  ldc2            0x00000000006d0d37 Type::getTypeInfo(Scope*) + 471
7  ldc2            0x000000000070793a DtoTypeInfoOf(Type*, bool) + 26
8  ldc2            0x00000000006fae55 RTTIBuilder::push_typeinfo(Type*) + 21
9  ldc2            0x00000000006d1dab TypeInfoConstDeclaration::llvmDefine() + 107
10 ldc2            0x00000000006d14d4 DtoDeclareTypeInfo(TypeInfoDeclaration*) + 244
11 ldc2            0x00000000006d16fe DtoResolveTypeInfo(TypeInfoDeclaration*) + 462
12 ldc2            0x00000000006d0d37 Type::getTypeInfo(Scope*) + 471
13 ldc2            0x000000000070793a DtoTypeInfoOf(Type*, bool) + 26
14 ldc2            0x00000000006fae55 RTTIBuilder::push_typeinfo(Type*) + 21
15 ldc2            0x00000000006d28f7 TypeInfoArrayDeclaration::llvmDefine() + 103
16 ldc2            0x00000000006d14d4 DtoDeclareTypeInfo(TypeInfoDeclaration*) + 244
17 ldc2            0x00000000006d16fe DtoResolveTypeInfo(TypeInfoDeclaration*) + 462
18 ldc2            0x00000000006d0d37 Type::getTypeInfo(Scope*) + 471
19 ldc2            0x000000000070793a DtoTypeInfoOf(Type*, bool) + 26
20 ldc2            0x000000000070038f
21 ldc2            0x000000000070079a DtoArrayEquals(Loc&, TOK, DValue*, DValue*) + 58
22 ldc2            0x00000000006ed678 EqualExp::toElem(IRState*) + 936
23 ldc2            0x00000000006e85f7 Expression::toElemDtor(IRState*) + 535
24 ldc2            0x00000000006dd210 ReturnStatement::toIR(IRState*) + 848
25 ldc2            0x00000000006dce99 CompoundStatement::toIR(IRState*) + 89
26 ldc2            0x00000000006b67e9 DtoDefineFunction(FuncDeclaration*) + 1641
27 ldc2            0x00000000007145a9 Ir::emitFunctionBodies() + 41
28 ldc2            0x00000000006a7d95 Module::genLLVMModule(llvm::LLVMContext&, Ir*) + 709
29 ldc2            0x00000000005a69c8 main + 7160
30 libc.so.6       0x00007f2487713ea5 __libc_start_main + 245
31 ldc2            0x00000000005bc7c9
Error: Error executing /home/jonathan/ldc/build/bin/ldc2: Segmentation fault (core dumped)

[redstar]

Yes, it crashes. Thanks for the report and the nice test case.

[dnadlinger]

The issue might be related to the fact that InnerStruct seems to be associated with multiple Type instances (the check in DtoType trips).

[dnadlinger]

The mangled name of the return type template changes from S14ldc_github_4704test18__T10TestStructTiZ10TestStruct to S14ldc_github_4704testFZS14ldc_github_4704test18__T10TestStructTiZ10TestStruct18__T10TestStructTiZ10TestStruct, which breaks the type merging after stripModifiers in DtoType. Sigh.

[dnadlinger]

Might be worth revisiting Kenji's solution to the similar issues we previously had with inout Voldemort functions for context.

[dnadlinger]

Still crashes.

[dnadlinger]

Well, turns out that the change in symbol codegen order that came with my recent changes mitigated this so that a tiny extra check in one place is enough to handle the borked AST.