fuzzing: compilation error ‘__builtin_mul_overflow’ does not have pointer to integer type #18389

issue naumenkogs opened this issue on March 20, 2020

naumenkogs commented at 12:29 AM on March 20, 2020: member

I'm using Debian, compiling current master (5bf45fe2a9642f8ae8f8a12bcbf8f8b4770421ad) after configure with ./configure --disable-shared --enable-tests --enable-fuzz CC=${AFLPATH}/afl-clang CXX=${AFLPATH}/afl-clang++.

Getting this error:

  CXX      test/fuzz/test_fuzz_key-fuzz.o
afl-cc 2.52b by <lcamtuf@google.com>
test/fuzz/integer.cpp:152:27: error: default initialization of an object of const type 'const SigningProvider' without a
      user-provided default constructor
    const SigningProvider store;
                          ^
                               {}
1 error generated.
Makefile:14931: recipe for target 'test/fuzz/test_fuzz_integer-integer.o' failed

If I configure for gcc, I get this error:

test/fuzz/multiplication_overflow.cpp: In instantiation of ‘void {anonymous}::TestMultiplicationOverflow(FuzzedDataProvider&) [with T = bool]’:
test/fuzz/multiplication_overflow.cpp:41:58:   required from here
test/fuzz/multiplication_overflow.cpp:21:75: error: argument 3 in call to function ‘__builtin_mul_overflow’ does not have pointer to integer type
     const bool is_multiplication_overflow_builtin = __builtin_mul_overflow(i, j, &result_builtin);
                                                                           ^
Makefile:15155: recipe for target 'test/fuzz/test_fuzz_multiplication_overflow-multiplication_overflow.o' failed

Funny enough, it works on my macbook locally :) cc @practicalswift

naumenkogs added the label Bug on Mar 20, 2020
fanquake added the label Tests on Mar 20, 2020
MarcoFalke commented at 12:55 PM on March 20, 2020: member

cc @practicalswift #18353 (review)
practicalswift commented at 3:36 PM on March 20, 2020: contributor

Thanks for the ping! I'll fix :)
practicalswift commented at 4:20 PM on March 20, 2020: contributor

@naumenkogs Thanks for reporting. The __builtin_mul_overflow issue is fixed by #18393.

I cannot reproduce the "default initialization of an object of const type …" issue. What versions of AFL and gcc are you using? :)
MarcoFalke commented at 4:56 PM on March 20, 2020: member

I cannot reproduce the "default initialization of an object of const type …" issue. What versions of AFL and gcc are you using? :)

If I had to guess, it is probably an older version of gcc, like 4.8 or 5.3
naumenkogs commented at 4:58 PM on March 20, 2020: member

I cannot reproduce the "default initialization of an object of const type …" issue. What versions of AFL and gcc are you using? :)

This issue happens with clang: clang version 3.8.0-2ubuntu4 (tags/RELEASE_380/final) AFL I'm using is 2.52b
practicalswift commented at 8:40 AM on March 21, 2020: contributor

@naumenkogs Are you able to reproduce it under a newer version of Clang? :)

I'm usually running Clang 10 compiled from trunk but I now tested against Clang 4 which is the oldest version I have around. I'm unable to reproduce under Clang 4.
MarcoFalke commented at 1:15 PM on March 21, 2020: member

@practicalswift I think @naumenkogs is using debian stretch: https://packages.debian.org/stretch/clang
naumenkogs commented at 5:51 PM on March 21, 2020: member

After updating clang, the clang issue got fixed. Should this be mentioned in docs? I think there's no such requirement for Bitcoin Core.

As for the gcc issue — this still applies #18393 even after compilers update, waiting for the solution in that PR :)
naumenkogs commented at 6:10 PM on March 22, 2020: member

Fixed in 7c1ac70
MarcoFalke referenced this in commit 97b0687501 on Mar 23, 2020
MarcoFalke closed this on Mar 23, 2020
DrahtBot locked this on Feb 15, 2022

Contributors

Labels

Linked (view graph)

#18393 tests: Don't assume presence of __builtin_mul_overflow(…) in MultiplicationOverflow(…) fuzzing harness