Erroneous underscores in templates with templates

Describe the bug Analyzing a program with debug info (both dwarf and pdb) that contains a symbol with a template containing a template will apply a symbol which does not match the symbol in the debug info. The GnuDemangler seems to be the only thing to apply it correctly. The ms demangler gets this even more wrong and adds in things like “class_” and “struct_” into the symbol name.

To Reproduce Steps to reproduce the behavior:

1. Compile the code under attachments with optimizations disabled to prevent inlining.
2. Go to the main function and see Holder<Holder<int>_>::Holder instead of Holder<Holder<int>>::Holder

Expected behavior There should obviously be no underscored where there were none to begin with.

Attachments

template <typename T>
class Holder {
    T t;
public:
    Holder() : t(){}
    ~Holder() = default;
    T get() const { return t; }
};

int main() {
    Holder<Holder<int>> var;
    return var.get().get();
}

Environment (please complete the following information):

• OS: Microsoft Windows [Version 10.0.22000.556]
• Java Version: 11
• Ghidra Version: 10.1.2 and 10.2_DEV 21aa55cf2dd173c668ca042cf8d352fb8462394c

Additional context Sometimes a demangler will insert a space for readability. Removing it instead of replacing it with an underscore would be correct.

[astrelsky]

@ryanmkurtz this also occurs with DWARF import and PDB loading.

[astrelsky]

Yes. The “known issue” is “an error,” and that is the subject code that replaces the characters, but supposedly to ensure the underlying problems do not get triggered.

I make the exact same arguments to others on the team; I feel that someone or people on the team who did not know languages well made that decision, and that is exactly what I was repeating to someone on the team yesterday.

It is related to your #1175.

I’d like to fix it or have others fix it, but there are some underlying issues that are being worked on that have bearing on this issue.

That’s going pretty far back now. I mistook your initial response as a won’t fix and am willing to put up a fight about it. It may seem like a stupid and minor problem but when having to manually rename 100s of labels so that the correct this type gets used, as well as so it is correct, gets painfully tedious.