Experiment with optional mode where sym files contain mangled names

[Gankra]

Filing this so I don't forget it. This could significantly improve compression, and actually help downstream code in understanding the "structure" of the names as the mangling includes some metadata that you have to moreso guess at from the demangled version.

Apparently it's as simple as removing all the name.demangle calls in these files:

dump_syms/src/linux/elf.rs

Lines 119 to 151 in 213bd65

	fn demangle(name: &Name) -> String {
	let name = common::fix_symbol_name(name);
	if let Language::C = name.language() {
	return name.as_str().to_string();
	}
	match name.demangle(DemangleOptions::complete()) {
	Some(demangled) => demangled,
	None => {
	let aname = name.as_str();
	warn!("Didn't manage to demangle {:?}", name);
	aname.to_string()
	}
	}
	}
	fn demangle_str(name: &str) -> String {
	let lang = Name::new(name, NameMangling::Mangled, Language::Unknown).detect_language();
	if lang == Language::Unknown {
	return name.to_string();
	}
	let name = Name::new(name, NameMangling::Mangled, lang);
	let name = common::fix_symbol_name(&name);
	match name.demangle(DemangleOptions::complete()) {
	Some(demangled) => demangled,
	None => {
	warn!("Didn't manage to demangle {}", name);
	name.to_string()
	}
	}
	}

dump_syms/src/windows/symbol.rs

Line 573 in eb4e203

match name.demangle(DemangleOptions::complete()) {

---

Rough design sketch:

• Off by default, enabled by passing --keep-mangled to dump_syms
• If enabled, emit INFO MANGLED 1 at the ~top of the file to make it unambiguous to tools that it support it whether the names are mangled or not.

That's it. All the hard work would be in all the downstream tools checking for INFO MANGLED and embedding/invoking a demangler.

[Gankra]

initial conclusion: v0 mangled names are on average longer than the demangled name -- doing this makes the sym file longer. However this is without inlinees, which are presumably the home of really nasty generic balls.