Bugs reported and fixed

MIPT-V is an educational project, so we always want to teach most modern technologies and tools. However, they often contain bugs or some features are not implemented. That's why we report them in our nearest convenience.

Compiler and tools bugs reported

LLVM

All bugs are found and reported by Pavel Kryukov, unless mentioned otherwise.

• Clang++ bugs:

  • llvm-project/llvm#35632 — False positive detection of unneeded declaration for user-defined literals used inside class templates. Bug observed by Aleksandr Misevich while debugging MIPT-MIPS/#276

• libc++ bugs:

  • llvm-project/llvm#34402 ­— missing 'noexcept' specifier for string_view literals (check-in)

• Static Analyzer false positives and UX enhancements:

  • llvm-project/llvm#35631 — False positive dead code detection for C++17 structured binding
  • llvm-project/llvm#43410 — cplusplus.NewDeleteLeaks should not report on system headers
  • llvm-project/llvm#44859 — NewDeleteLeaks reports false positive with shared_ptr

• Clang-Tidy false positives and UX enhancements:

  • llvm-project/llvm#34805 — „possible exception“ in constant expression
  • llvm-project/llvm#39117 — fuchsia-multiple-inheritance should not warn on class forward declaration.
  • llvm-project/llvm#39945 — do not try to reinvent a constexpr version of #define COMMA ,.
  • llvm-project/llvm#39946 — do not try to reinvent a constexpr version of #define STRING(x) # x.
  • llvm-project/llvm#39947 — support regexp for NOLINT statements.
  • llvm-project/llvm#33950 — int main() cannot have std::array as an argument.
  • llvm-project/llvm#39979 — std::string is not a magic number.
  • llvm-project/llvm#39986 — template instantiations should not expose enums as 'magic numbers'.
  • llvm-project/llvm#42454 — do not warn about ODR violation for constexpr template variable.
  • llvm-project/llvm#43376 — fuchsia-multiple-inheritance should not warn on template class instantiation.
  • llvm-project/llvm#43574 — bugprone-branch-clone false positive with template functions and constexpr
  • llvm-project/llvm#44161 — bugprone-infinite-loop false positive with CATCH2
  • llvm-project/llvm#44852 — readability-simplify-boolean-expr should not trigger on if statement with initializer
  • llvm-project/llvm#45580 — hicpp-use-nullptr does not make sense with spaceship operator
  • llvm-project/llvm#53961 — hicpp-use-nullptr does not make sense with spaceship operator (reopened)
  • llvm-project/llvm#53997 — cppcoreguidelines-pro-bounds-constant-array-index can be smarter for std::array
  • llvm-project/llvm#54011 — misc-redundant-expression false positive with operator== overload

• Clang-Tidy hicpp-signed-bitwise has a special attention as we do lots of shifts

  • llvm-project/llvm#36311 — bitwise assignment operators (&=, |=, ^=, <<=, >>=) should be analyzed
  • llvm-project/llvm#36309 — do not blame 1u << 2 for containing a signed integer literal.
  • llvm-project/llvm#44322 — hicpp-signed-bitwise.IgnorePositiveIntegerLiterals should be recursive
  • llvm-project/llvm#44391 — hicpp-signed-bitwise false positive with byte swapping code

Visual Studio

• Bug 225040 — False positive on UB right shifting in unreachable code.
• Bug 451137 — internal compiler error while referencing template variable
• Bug 457095 — cannot bind temporary int object to an rvalue reference
• Bug 457098 — extern template instantiation does not work for virtual functions
• Bug 833637 — ones-propagating right shift is optimized incorrectly. Firstly observed by Maxim Davydov in MIPT-MIPS/#1143
• Bug 846216 — false positive warning C4297 for constructor not a bug
• Bug 1662465 — W4100 with concept parameter

LGTM.com/CodeQL

LGTM.com is a great tool to detect possible errors. We report false positives:

• github/codeql#5709 C++ std::make_unique is considered as a stack variable.

Other

• We used to build GNU Binutils with more pedantic Clang compiler as a part of continuous integration flow. During the procedure while resolving MIPT-MIPS/#147, a bug in GNU Binutils was found by Kirill Nedostoev. Methodology was extracted to a separate repository to find more bugs in Binutils and Clang.
• PR libstdc++/82685 — missing 'noexcept' specifier for string_view literals

External pull requests

• We improve RISC-V community infrastructure:
  • riscv/riscv-opcodes#23
  • riscv/riscv-opcodes#26 by Rustem Yunusov
  • riscv/riscv-tests#183
  • riscv/riscv-opcodes#41
  • riscv/riscv-opcodes#53
  • riscv/riscv-opcodes#112
  • riscv/riscv-opcodes#113
  • riscv/riscv-opcodes#114
  • riscv/riscv-opcodes#118
  • riscv/riscv-opcodes#119
  • riscv/riscv-opcodes#120
  • riscv/riscv-opcodes#121
  • riscv/riscv-opcodes#122
• Enhancements to Boost C++ libraries:
  • boostorg/multiprecision#130
• We contribute to POPL option parser library, listed in POPL 1.2.0 release note.
• Minor fixes for ELFIO: serge1/ELFIO#26, serge1/ELFIO#27, serge1/ELFIO#28.
• Patches for GDB to support C++ simulators:
  • 2283a21 Change "xor" name in cpu_core to allow building with iso646.h or C++ compiler by Vyacheslav Kompan
  • 444b3fa Add 'extern C' if simulator is written in C++
  • 8f0a214 Include <string.h> to dis-asm.h to get strchr declaration
  • dc7e818 sim: Fix definition of SIM_ARANGE_INLINE
  • 7516c26 Do not expand macros to 'defined'
  • f47674b sim-utils.c: prevent buffer overflow.
• Infrastructure improvements for CEN64:
  • https://github.com/n64dev/cen64/pull/100
  • https://github.com/n64dev/cen64/pull/101
  • https://github.com/n64dev/cen64/pull/102
  • https://github.com/n64dev/cen64/pull/104
  • https://github.com/n64dev/cen64/pull/105
  • https://github.com/n64dev/cen64/pull/107
  • https://github.com/n64dev/cen64/pull/108
  • https://github.com/n64dev/cen64/pull/109
  • https://github.com/n64dev/cen64/pull/110
  • https://github.com/n64dev/cen64/pull/137
  • https://github.com/n64dev/cen64/pull/141
  • https://github.com/n64dev/cen64/pull/142
  • https://github.com/n64dev/cen64/pull/143
  • https://github.com/n64dev/cen64/pull/144
• Fix for MS Visual Studio build of LibELF was reported to Michael "Tired" Riepe who created the library. Unfortunately, Michael had not responded yet, so changes were published to WolfgangSt/libelf#1.