On Thu, Aug 06, 2020 at 09:29:59PM +0200, Luc Van Oostenryck wrote: > Sparse warns shifts by a negative or oversized amount but > it does this even on code that will not be executed. It's > annoying because such warnings are given for generic macros. > > The strategy for these warnings is changed in patch 4: > they are delayed until after the elimination of deadcode. > This uncovered a bug in the type evaluation and the linearization > of shift-assigns which is now solved in patch 2 & 3. > > Thanks to Stafford Horne to bring this back to my attention. > > This series is available for testing & review at: > git://git.kernel.org/pub/scm/devel/sparse/sparse.git bad-shift-equal Thanks! I see you merged this already, I tested it and see no regressions on my kernel build, and confirm the xchg issues are fixed. Sorry, I reviewed this early but didn't respond until now as I had time to test. I just had one question on 4/4. -Stafford > Luc Van Oostenryck (4): > shift-assign: add more testcases for bogus linearization > shift-assign: fix linearization of shift-assign > shift-assign: restrict shift count to unsigned int > bad-shift: wait dead code elimination to warn about bad shifts > > evaluate.c | 11 +- > expand.c | 18 -- > linearize.c | 44 ++++ > simplify.c | 20 +- > validation/expand/bad-shift.c | 8 +- > validation/linear/bug-assign-op0.c | 1 - > validation/linear/shift-assign1.c | 319 +++++++++++++++++++++++++++++ > validation/linear/shift-assign2.c | 53 +++++ > validation/optim/shift-big.c | 12 +- > validation/shift-negative.c | 4 +- > validation/shift-undef-long.c | 7 +- > validation/shift-undef.c | 52 ++--- > 12 files changed, 462 insertions(+), 87 deletions(-) > create mode 100644 validation/linear/shift-assign1.c > create mode 100644 validation/linear/shift-assign2.c > > > base-commit: 4c6cbe557c48205f9b3d2aae4c166cd66446b240 > -- > 2.28.0 >
