Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* first cut at reducing warnings binaryfusefilter.h only issues addressed 1. changed some return value and parameter types of (static) functions -- PLEASE CHECK THIS IN REVIEW 2. sprinkled 'U' into bitwise operations to silence warnings 3. casting to avoid "standard integer promotion rules" which resulted in signedness warnings 4. explicitly reducing results to the target type rather than letting it happen implicitly tests still passing * first cut at reducing warnings binaryfusefilter.h only issues addressed 1. changed some return value and parameter types of (static) functions -- PLEASE CHECK THIS IN REVIEW 2. sprinkled 'U' into bitwise operations to silence warnings 3. casting to avoid "standard integer promotion rules" which resulted in signedness warnings 4. explicitly reducing results to the target type rather than letting it happen implicitly 5. when and `if` statements ends in break or return, then a following `else if` can be just a new `if` tests still passing * starting work on xofilter.h * binclude/binaryfusefilter.h apparently clean for first time * formatting * first cut on xofilter.h mostly casting size_t down to uint32_t - maybe some internal struct types should have been size_t? also some integer promotion casts * round2 on xorfilter.h mostly casting blocklengt to uint32_t to fit into keyindex.index should keyindex.index be a size_t? * bench.c and unit.c very repetitive casting of mainly sizes and doubles. * all silent now on a clean compile with -Wconversion and -Wsign-conversion so putting these in the Makefile, so during "private" development with the Makefile new warnings will be noticed straight away but not in CMakeLists.txt, because as this is a header-only INTERFACE library, it would force these warning levels on the users. * another sweep from including c++ project turned up these additional 'U' tweaks * mistaken cast which broke test * factoring out the report functionality all sections were indentical except for the call to *contain() and *size_in_bytes some void* and function pointer juggling allowed to make this generic report code reduced by 2/3rds * iron out slight inconsistencies between tests * abstracting away the rest of the test logic for all but the special "failure rate test" the large function dispatch table is a litle annoying, but can be removed as well...TBC tests all pass * fixing a memory leak caught by sanitizer just a missing free() * _duplicates test cases can be convered by the same code remove initialization. it's not needed and compiler now happy * removing the need for large array of boiler plate function wrappers instead of having a wrapper function per action per filter type, we can cast the functions as a generic function pointer on the into the generic test runner, and then cast them as a compatible function pointer type inside the test runner. The generic `filter*` parameter cannot be `void*` and must be a dummy struct because: § 6.2.5..28: "All pointers to structure types shall have the same representation and alignment requirements as each other" (the same is not true for `void*` which may have different representation. This simple change results in a large code reduction and removes the unsightly and hard to remove array of boiler plate function wrappers. * upgrade CI use ubuntu 24.04 run a matrix with both gcc and clang compile tests/unit target with full warnings use -Werror so any new warnings are kept out compile the test with sanitzers to catch runtime UBSAN/ASAN issues execute ctest run with env options to ensure UBSAN/ASAN warning result in CI failure * change options only * correct compile and link options for test this CI build should now fail with UBSAN warning on clang (but would pass on gcc, and would pass on ubuntu 22.04 clang) * syntax * remove whitespce after \ * use sanitizers only on *nix not on mingw as not supported * fix the UBSAN warning this was the motivaing origin of all the CI changes resolve the UBSAN warning due to recent LLVM changes in clang-17 UBSAN, which ap0plies much more strict interpretation of casting of function pointers where one of the parameters is a void* and therefore not "an exact match" for the function ultimately called. this is not a problem in practice at runtime as these pointers are binary compatible, but best to avoid the interpretation of the standard by clang UBSAN - as major projects have also had to do The warning is only generated at runtime with sanatizers from recent clang-17+ compiled in. Hence the fairly wide reaching changes to CI. Solution to the UBSAN warning is macro generated "thunks", ie simple function wrappers, which allow the compiler to be happy about the cast. Much more terse than spelling out the wrapper functions, but with all the benefits of conformance and clangd/IDE help with completion and type checks. * really skip saninitizers unless not mingw * fixing warnings on MINGW32 immediately caught by the new -Werror policy, MINGW32 CI build was failing because: - size_t is 32bit on MINGW32 - filter->blockLength is uint64_t so any calculation involving filter->blockLength and being stored in a size_t was causing an implicit cast which this commit makes explicit review whether to just make filter->blockLength a size_t so it auto adjusts to the correct size.
- Loading branch information
1 parent
872a8bb
commit 9f5b776