ubsan in unit/fp/fadd.opt: shift exponent 111 is too large

[katrinafyi]

When running ubsan, there is an error reported in the fadd.opt unit test.

I only see this when using clang++ as the compiler and with the -fno-sanitize-recover=undefined flag added. Using g++ with the same CXXFLAGS does not seem to show the error.

export CXXFLAGS="-fsanitize=address,undefined -fno-sanitize-recover=undefined"
export CXX=$(which clang++)
rm -rf build; cmake -B build -GNinja -DCMAKE_PREFIX_PATH=~/progs/llvm-regehr/build/ -DBUILD_TV=1 -DCMAKE_VERBOSE_MAKEFILE=TRUE . && cmake --build build -j4
bash -c 'cd build; ../tests/lit/lit.py  ../tests/unit/fp/fadd.opt  -j10 --verbose'

-- Testing: 1 tests, 1 workers --
FAIL: Alive2 :: unit/fp/fadd.opt (1 of 1)
******************** TEST 'Alive2 :: unit/fp/fadd.opt' FAILED ********************
Processing /home/rina/progs/alive2/tests/unit/fp/fadd.opt..

----------------------------------------
Name: fsub, double, type infer
  %t = fsub double %a, 0.0
  ret double %t
=>
  %t = double %a
  ret %t

Done: 1
Transformation seems to be correct!

----------------------------------------
Name: fadd identical, double, constant
  %t = fadd double %a, 0.0
  ret double %t
=>
  %t = fadd double %a, 0.0
  ret double %t

Done: 1/home/rina/progs/alive2/ir/type.cpp:464:31: runtime error: shift exponent 111 is too large for 64-bit type 'unsigned long long'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior /home/rina/progs/alive2/ir/type.cpp:464:31 

This is 332cbac and clang version:

clang version 19.1.5
Target: x86_64-suse-linux

I noticed that https://github.com/regehr/alive2/blob/master/.github/workflows/build.yml is meant to run the tests with sanitizers, but I think that the environment variables should be named CXX and CXXFLAGS instead of their CMAKE_ versions. Also, -fno-sanitize-recover is needed, otherwise it only prints the violation and continues the program.

[regehr]

do you still get this with Alive2's top of tree, Kait? Nuno fixed a bug like this that someone else recently reported

[regehr]

or I guess he just merged an external fix!
9b2d131

[katrinafyi]

I just checked and it does seem to happen on 332cbac, after that commit.