donat.nagy added a comment.
In D158499#4608974 <https://reviews.llvm.org/D158499#4608974>, @danix800 wrote:
> One of the observable issue with inconsistent size type is
>
> void clang_analyzer_eval(int);
>
> typedef unsigned long long size_t;
> void *malloc(unsigned long size);
> void free(void *);
>
> void symbolic_longlong_and_int0(long long len) {
> char *a = malloc(5);
> (void)a[len + 1]; // no-warning
> // len: [-1,3]
> clang_analyzer_eval(-1 <= len && len <= 3); // expected-warning {{TRUE}}
> clang_analyzer_eval(0 <= len); // expected-warning
> {{UNKNOWN}}
> clang_analyzer_eval(len <= 2); // expected-warning
> {{UNKNOWN}}
> free(a);
> }
>
> which is extracted from
> `clang/test/Analysis/array-bound-v2-constraint-check.c`,
> with `DynamicMemoryModeling` turned on,
> the second warning does not hold anymore: `clang_analyzer_eval(0 <= len);`
> will be reported as `TRUE` which is not expected.
>
> `DynamicMemoryModeling` will record the extent of allocated memory as `5ULL`,
> `ArrayBoundV2` will do `len + 1 < 5ULL` assumption, simplified to `len <
> 4ULL`,
> which casts `len` to unsigned, dropping `-1`, similar to
>
> void clang_analyzer_eval(int);
>
> void test(int len) {
> if (len >= -1 && len <= 4U) { // len is promoted into unsigned, thus can
> never be negative
> clang_analyzer_eval(0 <= len); // TRUE
> }
> }
This issue is very interesting and scary -- I wouldn't have guessed that the
core constraint handling algorithms are so buggy that things like this can
happen 😦. I reproduced the issue and double-checked that the logic error is
//not// in ArrayBoundV2 (despite that its simplification algorithm is messy and
suspicious); I'd be very happy if this issue could be resolved (perhaps along
with other inaccuracies of APSInt math...).
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D158499/new/
https://reviews.llvm.org/D158499
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
