Rework script

3 files changed, 728 insertions, 74 deletions

diff --git a/tools/ghidra_scripts/DecompMapToGhidra.py b/tools/ghidra_scripts/DecompMapToGhidra.py
index abcc1aa2..053edb14 100644
--- a/tools/ghidra_scripts/DecompMapToGhidra.py
+++ b/tools/ghidra_scripts/DecompMapToGhidra.py
@@ -1,5 +1,4 @@
 #Take symbols from the decomp's map files, treat them as authoritative, and import them in Ghidra.
-#Requires cwdemangle in the PATH
 #Must run `python .\configure.py --map && ninja` beforehand!
 #@author robojumper
 #@category GameCube/Wii
@@ -9,7 +8,9 @@
 
 import os
 import re
-import subprocess
+
+import demangle
+demangle.mode = 'demangle'
 
 from ghidra.app.util import NamespaceUtils
 from ghidra.program.model.symbol import SymbolUtilities
@@ -32,22 +33,75 @@ allowed_sections = [
     '.rodata',
 ]
 
-def demangle(name):
+commit = None
+
+
+def postprocess_demangled_name(demangled):
+    if demangled.startswith("vtable for "):
+        demangled = demangled[len("vtable for "):] + "::__vtable"
+
+    if demangled.endswith(" const"):
+        demangled = demangled[:-len("const ")]
+
+    thunk = False
+    guard = False
+    if demangled.startswith("non-virtual thunk to "):
+        thunk = True
+        demangled = demangled[len("non-virtual thunk to "):]
+
+    if demangled.startswith("virtual thunk to "):
+        thunk = True
+        demangled = demangled[len("virtual thunk to "):]
+
+    if demangled.startswith("guard variable for "):
+        guard = True
+        demangled = demangled[len("guard variable for "):]
+
+    template_open = demangled.index("<") if "<" in demangled else None
+    first_space = demangled.index(" ") if " " in demangled else None
+    open_paren = demangled.index("(") if "(" in demangled else None
+    if template_open and first_space and open_paren and first_space < template_open and template_open < open_paren:
+        # this looks like a templated return type, so drop the return type
+        demangled = demangled[(first_space+1):]
+
+    demangled = demangled.replace("(anonymous namespace)", "anonymous")
+    demangled = demangled.replace("operator ", "operator_")
+    demangled = demangled.replace(" ", "")
+    if ")::" in demangled:
+        # dFontMng_c::getFontPath(unsigned char)::TEMP_FONT_NAME
+        left = demangled.split("(")[0]
+        right = demangled.rsplit(")")[-1]
+        # dFontMng_c::getFontPath::TEMP_FONT_NAME
+        demangled = left + right
+    else:
+        demangled = demangled.split("(")[0]
+
+    if thunk:
+        demangled += "_thunk"
+    if guard:
+        demangled += "_guard"
+
+    return demangled
+
+
+def do_demangle(name):
     # try demangling
     if "__" in name:
         try:
-            output = subprocess.check_output(["cwdemangle", name], stderr=subprocess.STDOUT)
-            return output.strip().split("(")[0]
-        except subprocess.CalledProcessError as e:
-            if "Failed to demangle symbol" not in e.output:
-                raise
+            output = demangle.demangle_try(name)
+            return output.strip()
+        except Exception:
+            pass
     # otherwise we try to undo the effects of the original
     # ghidra -> symbols.txt export here
-    if not "$" in name and not "arraydtor" in name and not name.startswith("__"):
+    if "$" not in name and "arraydtor" not in name and not name.startswith("__"):
         name = name.replace("__", "::")
         name = name.replace("::::", "::__")
     return name
 
+
+default_sym_re = re.compile(".*_[0-9A-Fa-f]{8}$")
+
 def parse_symbol(line):
     if "entry of" in line:
         return None
@@ -65,7 +119,6 @@ def parse_symbol(line):
         'vAddr': int(vAddr, 16),
     }
 
-default_sym_re = re.compile(".*_[0-9A-Za-z]{8}$")
 
 def parse_map_file(file):
     lines = [line for line in file]
@@ -77,7 +130,7 @@ def parse_map_file(file):
             section_name = line.split(' ')[0]
             if section_name in allowed_sections:
                 sections[section_name] = []
-                i += 4 # go to symbols
+                i += 4  # go to symbols
                 while lines[i].strip() != "":
                     sym = parse_symbol(lines[i])
                     if sym is not None:
@@ -87,93 +140,151 @@ def parse_map_file(file):
 
     return sections
 
+
+anon_static_re = re.compile("^@[0-9]+$")
+
+
+def symbol_needs_history(name):
+    if anon_static_re.match(name) or "arraydtor" in name:
+        return False
+
+    return True
+
+
 # This script works incrementally by recording
 # the mangled name in a special plate comment.
 # If the mangled name is the same, we don't even bother
 # shelling out to cwdemangle, speeding up the whole process
 # quite a bit.
-mangled_prefix = "mangled-decomp-name-v1: "
-def update_addr(addr, mangled_name):
+mangled_prefix = "mangled-name"
+
+# We also keep a history of the original Ghidra name and previous
+# decomp names in this plate comment, for future reference
+original_prefix = "original-name"
+previous_prefix = "full-name"
+
+
+def parse_comment(plate_comment):
+    ret = {
+        "other": [],
+        "original": None,
+        "history": [],
+        "mangled": None,
+    }
+
+    if plate_comment:
+        for line in plate_comment.splitlines():
+            if line.startswith(mangled_prefix):
+                ret["mangled"] = line.split(" ", 1)[1].strip()
+            elif line.startswith(original_prefix):
+                ret["original"] = line.split(" ", 1)[1].strip()
+            elif line.startswith(previous_prefix):
+                ret["history"].append(line.split(" ", 1)[1].strip())
+            else:
+                ret["other"].append(line.strip())
+    return ret
+
+
+def update_addr(addr, mangled_name, create_function=False):
     unit = listing.getCodeUnitAt(addr)
     if not unit:
         return
-    
-    new_comment_line = mangled_prefix + mangled_name
-
-    comment = unit.getComment(PLATE_COMMENT)
-    update_symbol = False
-    new_comment = None
-    if not comment:
-        # no plate comment here, add one and set symbol
-        update_symbol = True
-        new_comment = [new_comment_line]
+
+    comment_str = unit.getComment(PLATE_COMMENT)
+    comment = parse_comment(comment_str)
+
+    existing_symbol = getSymbolAt(addr)
+    existing_name = existing_symbol.getName(True) if existing_symbol else None
+
+    if comment["mangled"] and comment["mangled"] == mangled_name:
+        # skip updating
+        return
+
+    if not comment["mangled"] and not comment["original"] and existing_name and not default_sym_re.match(existing_name):
+        comment["original"] = existing_name
+
+    demangled_name = do_demangle(mangled_name)
+    postprocessed = postprocess_demangled_name(demangled_name)
+    comment["mangled"] = mangled_name
+    comment["history"].append(commit + " " + demangled_name)
+
+    complete_plate_comment = comment["other"]
+    if comment["original"]:
+        complete_plate_comment.append(original_prefix + " " + comment["original"])
+    for h in comment["history"]:
+        complete_plate_comment.append(previous_prefix + " " + h)
+    complete_plate_comment.append(mangled_prefix + " " + mangled_name)
+
+    complete_plate_comment = "\n".join(complete_plate_comment)
+
+    name_list = [SymbolUtilities.replaceInvalidChars(part, True) for part in postprocessed.split("::")]
+    symbol_str = name_list[-1]
+    namespace = None
+    if len(name_list) > 1:
+        namespace_str = "::".join(name_list[:-1])
+        namespace = NamespaceUtils.createNamespaceHierarchy(namespace_str, None, currentProgram, IMPORTED)
+
+    sym = getSymbolAt(addr)
+    if namespace is None:
+        namespace = currentProgram.getGlobalNamespace()
+    if sym:
+        sym.setNameAndNamespace(symbol_str, namespace, IMPORTED)
     else:
-        comment_lines = comment.splitlines()
-        if any (line.startswith(mangled_prefix) for line in comment_lines):
-            # replace with new mangled name
-            new_comment = []
-            for line in comment_lines:
-                if line.startswith(mangled_prefix) and new_comment_line not in line:
-                    update_symbol = True
-                    new_comment.append(new_comment_line)
-                else:
-                    new_comment.append(line)
-
-        else:
-            # existing plate comment without symbol, append
-            update_symbol = True
-            new_comment = [comment, new_comment_line]
-
-    if update_symbol:
-        new_comment = '\n'.join(new_comment)
-        demangled_name = demangle(mangled_name)
-
-        # print(new_comment, demangled_name)
-
-        name_list = [SymbolUtilities.replaceInvalidChars(part, True) for part in demangled_name.split("::")]
-        symbol_str = name_list[-1]
-        namespace = None
-        if len(name_list) > 1:
-            namespace_str = "::".join(name_list[:-1])
-            print(mangled_name)
-            namespace = NamespaceUtils.createNamespaceHierarchy(namespace_str, None, currentProgram, IMPORTED)
-
-        sym = getSymbolAt(addr)
-        if namespace is None:
-            namespace = currentProgram.getGlobalNamespace()
-        if sym:
-            sym.setNameAndNamespace(symbol_str, namespace, IMPORTED)
-        else:
-            createLabel(addr, symbol_str, namespace, True, IMPORTED)
-        unit.setComment(PLATE_COMMENT, new_comment)
-
-def apply_symbols_map(symbols_map, file_name):
+        createLabel(addr, symbol_str, namespace, True, IMPORTED)
+
+    if create_function:
+        createFunction(addr, postprocessed)
+
+    if symbol_needs_history(mangled_name):
+        unit.setComment(PLATE_COMMENT, complete_plate_comment)
+
+
+def get_section_names(file_name, build_dir):
+    splits_path = build_dir.replace("build", "config")
+    splits_path = os.path.join(splits_path, "rels", file_name, "splits.txt")
+    section_names = []
+    with open(splits_path, "rt") as file:
+        for line in file:
+            line = line.strip()
+            if line == "":
+                break
+            elif line == "Sections:":
+                continue
+            else:
+                section_names.append(line.split()[0])
+    return section_names
+
+
+def apply_symbols_map(symbols_map, file_name, build_dir):
+    if file_name != "MAIN":
+        section_names = get_section_names(file_name, build_dir)
+        blocks = mem.getBlocks()
+        blocks = [b for b in blocks if b.getName().startswith(file_name)]
+
     for section, syms in symbols_map.items():
         for sym in syms:
             if file_name == "MAIN":
                 # in the main dol, each symbol is loaded at a fixed address
                 addr_obj = AF.getAddress("0x%08X" % sym["vAddr"])
             else:
-                # REL sections can't be reliably identified
-                if section != ".text":
-                    continue
-                # in rels, every section is relocated indivdually, so treat
+                index = section_names.index(section)
+                block = blocks[index]
+                # in rels, every section is relocated individually, so treat
                 # this as an offset
-                block_name = file_name + "_" + section + "0"
-                block = mem.getBlock(block_name)
                 addr_obj = block.getStart().add(sym["vAddr"])
-            update_addr(addr_obj, sym["name"])
+            is_text = section == ".text"
+            update_addr(addr_obj, sym["name"], create_function=is_text)
 
 
 path = str(askDirectory("Program build directory (e.g. build/SOUE01)", "Import"))
-
+commit = askString("Commit hash for symbol history", "Confirm")
 new_contents = None
 main_symbols = os.path.join(path, "main.elf.MAP")
 symbols_map = None
 with open(main_symbols, "rt") as file:
     symbols_map = parse_map_file(file)
 
-apply_symbols_map(symbols_map, "MAIN")
+apply_symbols_map(symbols_map, "MAIN", path)
 
 for rel_name in os.listdir(path):
     if rel_name.endswith("NP"):
@@ -182,4 +293,4 @@ for rel_name in os.listdir(path):
         with open(rel_symbols, "rt") as file:
             symbols_map = parse_map_file(file)
 
-        apply_symbols_map(symbols_map, rel_name)
+        apply_symbols_map(symbols_map, rel_name, path)
diff --git a/tools/ghidra_scripts/GhidraToDtkSymbols.py b/tools/ghidra_scripts/GhidraToDtkSymbols.py
index a8bcd913..23301f9c 100644
--- a/tools/ghidra_scripts/GhidraToDtkSymbols.py
+++ b/tools/ghidra_scripts/GhidraToDtkSymbols.py
@@ -12,7 +12,7 @@ AF = currentProgram.getAddressFactory()
 mem = currentProgram.getMemory()
 listing = currentProgram.getListing()
 
-sym_re = re.compile("(?:lbl|fn|FUN|DAT)_[0-9A-Fa-f_]+ = \.([a-z0-9]+):0x([0-9A-Fa-f]{8})(.*)\n")
+sym_re = re.compile("(?:lbl|fn|FUN|DAT)_[0-9A-Fa-f_]+ = \\.([a-z0-9]+):0x([0-9A-Fa-f]{8})(.*)\n")
 default_sym_re = re.compile(".*_[0-9A-Za-z]+$")
 
 used_symbols = set()
diff --git a/tools/ghidra_scripts/demangle.py b/tools/ghidra_scripts/demangle.py
new file mode 100644
index 00000000..9b116b5d
--- /dev/null
+++ b/tools/ghidra_scripts/demangle.py
@@ -0,0 +1,543 @@
+# https://gist.github.com/RootCubed/8f8102fe6cf4ed79a45f1dfe23020a06
+
+# Demangler / Itanium remangler for the CodeWarrior ABI
+
+# Adapted from the NVIDIA demangler script by Ninji
+# See https://gist.github.com/RootCubed/9ebecf21eec344f10164cdfabbf0bb41 (Python)
+# and https://gist.github.com/RootCubed/d7e2629f4576059853505b7931ffd105 (C++)
+# for those scripts
+
+# Ported to Ghidra's Jython / Python 2.7 by robojumper
+
+import argparse
+import re
+import sys
+
+mode = None
+verbose = False
+
+def is_demangle():
+    return mode == 'demangle'
+
+"""
+The names of the types in the output
+First demangled, then remangled for Itanium
+"""
+names_mapping = {
+    'v': ('void', 'v'),
+    'b': ('bool', 'b'),
+    'c': ('char', 'c'),
+    's': ('short', 's'),
+    'i': ('int', 'i'),
+    'l': ('long', 'l'),
+    'x': ('long long', 'x'),
+    'Sc': ('signed char', 'a'),
+    'Uc': ('unsigned char', 'h'),
+    'Us': ('unsigned short', 't'),
+    'Ui': ('unsigned int', 'j'),
+    'Ul': ('unsigned long', 'm'),
+    'Ux': ('unsigned long long', 'y'),
+    'f': ('float', 'f'),
+    'd': ('double', 'd'),
+    'r': ('long double', 'e'),
+    'w': ('wchar_t', 'w'),
+    'e': ('...', 'z')
+}
+
+"""
+The names of the methods in the output
+First demangled, then remangled for Itanium
+"""
+method_mapping = {
+    '__dt':  ('~$CLS$', 'D0'),
+    '__ct':  ('$CLS$', 'C1'),
+    '__nw':  ('operator new', 'nw'),
+    '__nwa': ('operator new[]', 'na'),
+    '__dl':  ('operator delete', 'dl'),
+    '__dla': ('operator delete[]', 'da'),
+    '__pl':  ('operator+', 'pl'),
+    '__mi':  ('operator-', 'mi'),
+    '__ml':  ('operator*', 'ml'),
+    '__dv':  ('operator/', 'dv'),
+    '__md':  ('operator%', 'rm'),
+    '__er':  ('operator^', 'eo'),
+    '__ad':  ('operator&', 'an'),
+    '__or':  ('operator|', 'or'),
+    '__co':  ('operator~', 'co'),
+    '__nt':  ('operator!', 'nt'),
+    '__as':  ('operator=', 'aS'),
+    '__lt':  ('operator<', 'lt'),
+    '__gt':  ('operator>', 'gt'),
+    '__apl': ('operator+=', 'pL'),
+    '__ami': ('operator-=', 'mI'),
+    '__amu': ('operator*=', 'mL'),
+    '__adv': ('operator/=', 'dV'),
+    '__amd': ('operator%=', 'rM'),
+    '__aer': ('operator^=', 'eO'),
+    '__aad': ('operator&=', 'aN'),
+    '__aor': ('operator|=', 'oR'),
+    '__ls':  ('operator<<', 'ls'),
+    '__rs':  ('operator>>', 'rs'),
+    '__ars': ('operator>>=', 'rS'),
+    '__als': ('operator<<=', 'lS'),
+    '__eq':  ('operator==', 'eq'),
+    '__ne':  ('operator!=', 'ne'),
+    '__le':  ('operator<=', 'le'),
+    '__ge':  ('operator>=', 'ge'),
+    '__aa':  ('operator&&', 'aa'),
+    '__oo':  ('operator||', 'oo'),
+    '__pp':  ('operator++', 'pp'),
+    '__mm':  ('operator--', 'mm'),
+    '__cm':  ('operator,', 'cm'),
+    '__rm':  ('operator->*', 'pm'),
+    '__rf':  ('operator->', 'pt'),
+    '__cl':  ('operator()', 'cl'),
+    '__vc':  ('operator[]', 'ix'),
+}
+
+def parse_number(s, i):
+    """
+    Parses a number starting at position i.
+    Examples:
+        parse_number('123ABC', 0) -> (123, 3)
+
+    Args:
+        s (str): The input string to parse.
+        i (int): The starting position in the input string.
+
+    Returns:
+        Tuple[int, int]: The parsed number and the new position in the string.
+    """
+    num = 0
+    while s[i].isdigit():
+        num = num * 10 + int(s[i])
+        i += 1
+    return num, i
+
+def parse_typename(s, i):
+    """
+    Fully processes a mangled typename starting at index i.
+    Examples:
+        (demangle) parse_typename('Q23ABC3DEF', 0) -> ('ABC::DEF', 10)
+        (demangle) parse_typename('Q23ABC6DEF<c>', 0) -> ('ABC::DEF<char>', 13)
+        (remangle) parse_typename('Q23ABC3DEF', 0) -> ('3ABC3DEF', 10)
+
+    Args:
+        s (str): The input string.
+        i (int): The starting index.
+        is_toplevel (bool): Whether the type is a global-level type. Used for remangling.
+
+    Returns:
+        Tuple[str, int]:The parsed typename and the new position in the string.
+    """
+    if s[i] == 'Q':
+        count = int(s[i + 1])
+        i += 2
+        bits = []
+        for _ in range(count):
+            size, i = parse_number(s, i)
+            bits.append(resolve_templates(s[i:(i + size)], True))
+            i += size
+        if is_demangle():
+            return '::'.join(bits), i
+        else:
+            return ''.join(bits), i
+    else:
+        size, i = parse_number(s, i)
+        return resolve_templates(s[i:(i + size)], True), i + size
+    
+def join_modifiers(modifiers):
+    """
+    Joins the list of modifiers into a single string.
+    Modifiers are e.g. const, pointer, reference, etc.
+    In a demangled string these are right-to-left (e.g. int const * - pointer to const int)
+    whereas in a mangled string they are left-to-right (e.g. PKi - pointer to const int)
+
+    Args:
+        modifiers (list[str]): The list of modifiers to join.
+
+    Returns:
+        str: The joined string of modifiers.
+    """
+    if is_demangle():
+        return ''.join(modifiers[::-1])
+    else:
+        return ''.join(modifiers)
+
+def parse_function(s, i, modifiers, name='', rettype_mode = 'show'):
+    """
+    Parses a function from a demangled string.
+    Examples:
+        (demangle) parse_function('v_v', 0, ['*']) -> ('void (*) ()', 4)
+        (demangle) parse_function('s_b', 0, ['&']) -> ('bool (&) (short)', 4)
+        (remangle) parse_function('i_v', 0, ['*']) -> ('FviE', 4)
+
+    Args:
+        s (str): The demangled string.
+        i (int): The current index in the string.
+        modifiers (list[str]): The list of modifiers.
+        name (str): An identifier, if available. This is the "main" symbol name.
+        rettype_mode (Literal['show', 'hide_in_demangle', 'remove']): How to handle the return type.
+
+    Returns:
+        Tuple[str, int]: The transformed function signature and the new position in the string.
+    """
+    # Parse the function args, return type handled later
+    args = []
+    while i < len(s) and s[i] != '_' and s[i] != '@':
+        argtype, i = parse_type(s, i)
+        args.append(argtype)
+
+    # Special case: const
+    # Note that if the function is const, it will be the last modifier
+    # because e.g. CPFv is a (const pointer) to a function
+    const_str = ''
+    if len(modifiers) > 0 and (modifiers[-1] == ' const' or modifiers[-1] == 'K'):
+        const_str = ' const' if is_demangle() else 'K'
+        modifiers.pop()
+
+    mod_str = join_modifiers(modifiers)
+
+    if is_demangle():
+        if mod_str != '':
+            mod_str = '(%s)' % mod_str.strip()
+        arg_str = ', '.join(args) if args[0] != 'void' else ''
+        func_str = '%s%s(%s)%s' % (name, mod_str, arg_str, const_str)
+        if i >= len(s) or s[i] == '@':
+            return func_str, i
+        if rettype_mode == 'hide_in_demangle' or rettype_mode == 'remove':
+            _, i = parse_type(s, i + 1)
+            return func_str, i
+        else:
+            return parse_type(s, i + 1, [' ' + func_str])
+    else:
+        if i < len(s) and s[i] != '@':
+            rettype, i = parse_type(s, i + 1)
+        else:
+            rettype, i = ('', i)
+        if rettype_mode == 'remove':
+            rettype = ''
+        func_encoding = '%s%s' % (rettype, ''.join(args)) if name != '' else 'F%s%sE' % (rettype, ''.join(args))
+        if name != '':
+            func_encoding = 'N%s%sE%s' % (const_str, name, func_encoding)
+        else:
+            func_encoding = '%s%s' % (const_str, func_encoding)
+        return mod_str + func_encoding, i
+
+def parse_type(s, i, modifiers = None, name='', rettype_mode = 'show'):
+    """
+    Parses a type from a string - main transformation function.
+
+    Args:
+        s (str): The string to parse.
+        i (int): The starting index.
+
+    Returns:
+        Tuple[str, int]: The transformed type name and the new position in the string.
+    """
+
+    if modifiers == None:
+        modifiers = []
+
+    # Type modifier is for unsigned/signed
+    type_modifier = ''
+
+    while i < len(s) and s[i].isupper() and s[i] != 'Q':
+        c = s[i]
+        if c == 'C': # Const
+            modifiers.append(' const' if is_demangle() else 'K')
+        elif c == 'P': # Pointer
+            modifiers.append('*' if is_demangle() else 'P')
+        elif c == 'R': # Reference
+            modifiers.append('&' if is_demangle() else 'R')
+        elif c == 'V': # Volatile
+            modifiers.append(' volatile' if is_demangle() else 'V')
+        elif c == 'U' or c == 'S': # Unsigned/Signed
+            type_modifier = c
+        elif c == 'F': # Function, will return early
+            return parse_function(s, i + 1, modifiers, name, rettype_mode)
+        elif c == 'M': # Pointer-to-member
+            class_name, i = parse_type(s, i + 1)
+
+            modifiers.append(' %s::*' % class_name if is_demangle() else 'M%s' % class_name)
+            if s[i] == 'F':
+                # CW includes the hidden pointer arguments in the PTMF signature
+                # and also uses this to communicate constness of the PTMF
+                if s[i:].startswith('FPCvPCv'):
+                    modifiers.append(' const' if is_demangle() else 'K')
+                    i += 7
+                elif s[i:].startswith('FPCvPv'):
+                    i += 6
+                if s[i] == '_':
+                    # small hack: simulate Fv_... by reusing the v from FPCvPCv/FPCvPv
+                    i -= 1
+                return parse_function(s, i, modifiers)
+            else:
+                # pointer-to-member-nonfunction, continue parsing as normal
+                continue
+        elif c == 'A': # Array
+            count, i = parse_number(s, i + 1)
+            # Automatically skips past the '_' after the number before the next iteration
+            if is_demangle():
+                modstr = join_modifiers(modifiers)
+                if re.search(r'\[.*\]$', modstr) != None:
+                    modifiers.insert(0, '[%d]' % count)
+                elif modstr == '':
+                    modifiers.insert(0, ' [%d]' % count)
+                else:
+                    # modifiers.insert(0, f' ({modstr}) [{count}]')
+                    modifiers = [' (%s) [%d]' % (modstr, count)]
+            else:
+                modifiers.append('A%d_' % count)
+        else:
+            raise Exception('Invalid type modifier "' + c + '"')
+        i += 1
+
+    assert i < len(s)
+    assert s[i].isalpha() or s[i].isdigit()
+
+    # Now we have either an identifier or a basic type
+
+    if s[i] == 'Q' or s[i].isdigit():
+        type_name, i = parse_typename(s, i)
+        if not is_demangle():
+            type_name = 'N%sE' % type_name
+    else:
+        # Basic type - combine with type modifier and look up in mapping
+        actual_type = type_modifier + s[i]
+        if actual_type not in names_mapping:
+            raise Exception('Invalid type "' + actual_type + '"')
+        type_name = names_mapping[actual_type][0 if is_demangle() else 1]
+        i += 1
+
+    mod_str = join_modifiers(modifiers)
+
+    if is_demangle():
+        return '%s%s' % (type_name, mod_str), i
+    else:
+        return '%s%s' % (mod_str, type_name), i
+
+def resolve_templates(s, remangle_add_length):
+    """
+    Resolves template types in a type string.
+    Examples:
+        (demangle) resolve_templates('std<c>', false) -> 'std<char>'
+        (remangle) resolve_templates('std<c>', false) -> 'stdIcE'
+        (remangle) resolve_templates('std<c>', true) -> '3stdIcE'
+
+    Args:
+        s (str): The string to resolve.
+        remangle_add_length (bool): Whether to add the length prefix in remangling.
+
+    Returns:
+        str: The resolved string.
+    """
+    begin_pos = s.find('<')
+    if begin_pos == -1:
+        if re.match(r'^@unnamed@.+@$', s) != None:
+            if is_demangle():
+                # name.split('@')[2] contains the name of the file the anonymous namespace is in,
+                # but we lose that information here since we follow c++filt's behavior.
+                return '(anonymous namespace)'
+            else:
+                return '12_GLOBAL__N_1'
+        unnamed_type_m = re.match(r'^@class\$(\d*).+$', s)
+        if unnamed_type_m != None:
+            typenum = int(unnamed_type_m.group(1)) if unnamed_type_m.group(1) != '' else -1
+            if is_demangle():
+                return '{unnamed type#%d}' % (typenum + 2)
+            else:
+                return 'Ut%s_' % (str(typenum) if typenum > -1 else '')
+        if not is_demangle() and remangle_add_length:
+            return '%d%s' % (len(s), s)
+        return s
+    template_str = ''
+    i = begin_pos + 1
+    while i < len(s):
+        if s[i] == ',':
+            if is_demangle():
+                template_str += ', '
+            i += 1
+            continue
+        if s[i] == '>':
+            break
+        elif re.match(r'[-\d]+[>,]', s[i:]) != None:
+            # Integer literal
+            literal = re.match(r'[-\d]+', s[i:])[0]
+            template_str += literal if is_demangle() else 'XLi%sEE' % literal.replace('-', 'n')
+            i += len(literal)
+        else:
+            type, i = parse_type(s, i)
+            template_str += type
+    if is_demangle():
+        template_str = '<%s>' % template_str
+        # replicate c++filt behavior
+        if template_str[-2:] == '>>':
+            template_str = template_str[:-1] + ' >'
+        return s[0:begin_pos] + template_str
+    else:
+        if remangle_add_length:
+            return str(begin_pos) + s[0:begin_pos] + ('I%sE' % template_str)
+        return s[0:begin_pos] + ('I%sE' % template_str)
+
+def demangle(s):
+    """
+    Demangles a mangled symbol name.
+    """
+
+    at_sym = ''
+    thunk_offsets = []
+    m = re.match(r'^@([^@]+)@(.+)$', s)
+    if m != None:
+        m_thunk = re.match(r'^@(\d+)(?:@(\d+))?@(.+)$', s)
+        if m_thunk != None:
+            thunk_offsets = [int(m_thunk.group(1))]
+            if m_thunk.group(2) != None:
+                thunk_offsets.append(int(m_thunk.group(2)))
+            s = m_thunk.group(3)
+        else:
+            at_sym = m.group(1)
+            if at_sym not in ['LOCAL', 'GUARD', 'STRING']:
+                raise Exception('Invalid symbol name "' + s + '"')
+            s = m.group(2)
+
+    template_depth = 0
+    last_possible_end = -1
+    for i in range(1, len(s)):
+        if s[i] == '<':
+            template_depth += 1
+        elif s[i] == '>':
+            template_depth -= 1
+        if template_depth == 0 and i + 2 < len(s) and s[i:i + 2] == '__' and s[i + 2] in 'CFQ0123456789':
+            last_possible_end = i
+            break
+    if last_possible_end == -1:
+        return s
+
+    i = last_possible_end
+
+    method, remainder = s[:i], s[i + 2:]
+    if remainder[0] == 'F':
+        # Global function without class
+        class_name = ''
+        i = 0
+    else:
+        class_name, i = parse_typename(remainder, 0)
+
+    if '<' in method:
+        template_start = method.find('<')
+        pre_template, template = method[:template_start], method[template_start:]
+        resolved_templates = resolve_templates(template, False)
+    else:
+        pre_template, resolved_templates = method, ''
+
+    if pre_template in ['__ct', '__dt']:
+        rettype_mode = 'remove'
+    elif at_sym != '':
+        rettype_mode = 'hide_in_demangle'
+    else:
+        rettype_mode = 'show'
+
+    if method == '__vt':
+        return 'vtable for %s' % class_name if is_demangle() else '_ZTVN%sE' % class_name
+    elif method.startswith('__op'):
+        # Use method because the type might contain templates
+        cv_type_name, _ = parse_type(method[4:], 0)
+        pre_template = 'operator %s' % cv_type_name if is_demangle() else 'cv%s' % cv_type_name
+        # __op cannot be templated
+        resolved_templates = ''
+    elif pre_template in method_mapping:
+        pre_template = method_mapping[pre_template][0 if is_demangle() else 1]
+        if is_demangle():
+            # __ct should use the template of the function, not of the parent class
+            last_class_name = re.sub(r'<.+>', '', class_name).split('::')[-1]
+            pre_template = pre_template.replace('$CLS$', last_class_name)
+    else:
+        if not is_demangle():
+            pre_template = '%d%s' % (len(pre_template), pre_template)
+    
+    method = '%s%s' % (pre_template, resolved_templates)
+
+    if is_demangle():
+        demangled = '::'.join(filter(None, [class_name, method]))
+    else:
+        demangled = class_name + method
+
+    if i < len(remainder):
+        demangled, i = parse_type(remainder, i, name=demangled, rettype_mode=rettype_mode)
+    elif not is_demangle():
+        demangled = 'N%sE' % demangled
+
+    if i < len(remainder) and remainder[i] == '@' and at_sym in ['LOCAL', 'GUARD']:
+        subs = remainder[i + 1:].split('@')
+        local_sym_name = subs[0]
+        local_sym_extra = ('_' + subs[1]) if len(subs) > 1 else ''
+        if not is_demangle():
+            local_sym_name = str(len(local_sym_name)) + local_sym_name + local_sym_extra
+    elif at_sym == 'GUARD' and i >= len(remainder):
+        local_sym_name = method
+    elif at_sym == 'STRING' and i < len(remainder) and remainder[i] == '@' and not is_demangle():
+        local_sym_name = '_' + remainder[i + 1]
+    else:
+        local_sym_name = ''
+
+    if is_demangle():
+        if local_sym_name != '':
+            demangled += '::%s' % local_sym_name
+
+        # c++filt removes spaces in (* <symbol name>) -> (*<symbol name>), try to replicate this
+        while True:
+            m = re.search(r'\((?:[*&]|const| )+ (\w+.+)$', demangled)
+            if m == None or m.group(1).startswith('const'):
+                break
+            demangled = demangled[:m.start(1) - 1] + m.group(1)
+
+        if at_sym == 'GUARD':
+            return 'guard variable for %s' % demangled
+        elif at_sym == 'STRING':
+            return '%s::string literal' % demangled
+        elif len(thunk_offsets) > 0:
+            thunk_type = 'virtual' if len(thunk_offsets) == 2 else 'non-virtual'
+            return '%s thunk to %s' % (thunk_type, demangled)
+        else:
+            return demangled
+    else:
+        if len(thunk_offsets) == 1:
+            demangled = 'Th%d_%s' % (thunk_offsets[0], demangled)
+        elif len(thunk_offsets) == 2:
+            demangled = 'Tv%d_n%d_%s' % (thunk_offsets[0], thunk_offsets[1], demangled)
+        if at_sym == 'LOCAL':
+            demangled = 'Z%sE%s' % (demangled, local_sym_name)
+        if at_sym == 'GUARD':
+            demangled = 'GVZ%sE%s' % (demangled, local_sym_name)
+        if at_sym == 'STRING':
+            demangled = 'Z%sEs%s' % (demangled, local_sym_name)
+        return '_Z%s' % demangled
+
+def demangle_try(s):
+    try:
+        return demangle(s)
+    except Exception as e:
+        sys.stderr.write('Demangler error: ' + str(e) + '\n')
+        raise e
+
+def main():
+    global mode
+    global verbose
+    parser = argparse.ArgumentParser()
+    parser.add_argument('symbol', type=str, nargs='?')
+    parser.add_argument('-m', '--mode', choices=['demangle', 'remangle_itanium'], required=True)
+    parser.add_argument('-v', '--verbose', action='store_true', default=False)
+    args = parser.parse_args()
+    mode = args.mode
+    verbose = args.verbose
+    if args.symbol is None:
+        while True:
+            sym = input()
+            print(demangle_try(sym))
+    else:
+        print(demangle_try(args.symbol))
+        return
+
+if __name__ == '__main__':
+    main()