diff options
author | Ryan Prichard <rprichard@google.com> | 2024-02-23 16:43:48 -0800 |
---|---|---|
committer | Ryan Prichard <rprichard@google.com> | 2024-02-23 20:00:35 -0800 |
commit | 1a5e8713743270c2976a5f387e72ae0efbe646a7 (patch) | |
tree | 45447f6c98cb203055098dbf428a4933d0e4b79c | |
parent | d74f255a50d06bdeba1eefc64d80bcf51ac49b2c (diff) | |
download | bionic-1a5e8713743270c2976a5f387e72ae0efbe646a7.tar.gz |
cxa_demangle_test: expand accepted outputs
After updating libc++, the demangled output's float literal ends with
'L' and the <template-args> ends with a '>'. However, the input is
invalid, so the demangler probably should return nullptr.
Bug: http://b/175635923
Test: bionic-unit-tests
Change-Id: I8440118e4f5791a3464e15d6f9d2f5f3d006e54d
-rw-r--r-- | tests/__cxa_demangle_test.cpp | 30 |
1 files changed, 29 insertions, 1 deletions
diff --git a/tests/__cxa_demangle_test.cpp b/tests/__cxa_demangle_test.cpp index d400619ad..e13410c7f 100644 --- a/tests/__cxa_demangle_test.cpp +++ b/tests/__cxa_demangle_test.cpp @@ -28,11 +28,39 @@ #include <cxxabi.h> #include <gtest/gtest.h> +#include <string.h> TEST(__cxa_demangle, cxa_demangle_fuzz_152588929) { #if defined(__aarch64__) + // Test the C++ demangler on an invalid mangled string. libc++abi currently + // parses it like so: + // (1 "\006") (I (L e "eeEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE" E) E) + // There are a few interesting things about this mangled input: + // - The IA64 C++ ABI specifies that an FP literal's hex chars are lowercase. + // The libc++abi demangler currently accepts uppercase A-F digits, which is + // confusing because 'E' is supposed to mark the end of the <expr-primary>. + // - libc++abi uses snprintf("%a") which puts an unspecified number of bits + // in the digit before the decimal point. + // - The identifier name is "\006", and the IA64 C++ ABI spec is explicit + // about not specifying the encoding for characters outside of + // [_A-Za-z0-9]. + // - The 'e' type is documented as "long double, __float80", and in practice + // the length of the literal depends on the arch. For arm64, it is a + // 128-bit FP type encoded using 32 hex chars. The situation with x86-64 + // Android OTOH is messy because Clang uses 'g' for its 128-bit + // long double. char* p = abi::__cxa_demangle("1\006ILeeeEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE", 0, 0, 0); - ASSERT_STREQ("\x6<-0x1.cecececececececececececececep+11983", p); + if (p && !strcmp(p, "\x6<-0x1.cecececececececececececececep+11983")) { + // Prior to llvm.org/D77924, libc++abi left off the "L>" suffix. + } else if (p && !strcmp(p, "\x6<-0x1.cecececececececececececececep+11983L>")) { + // After llvm.org/D77924, the "L>" suffix is present. libc++abi + // accepts A-F digits but decodes each using (digit - 'a' + 10), turning 'E' + // into -18. + } else { + // TODO: Remove the other accepted outputs, because libc++abi probably + // should reject this input. + ASSERT_EQ(nullptr, p) << p; + } free(p); #endif } |