[Bug tree-optimization/119070] gcc15 incorrectly reporting negative array-bounds errors
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119070 --- Comment #4 from Taylor Hutt --- (In reply to Andrew Pinski from comment #3) > I don't see anything wrong with the warning since it is undefined in C to > what this testcase is trying to do. > > Do you have a full source rather than a reduced one? Because the reduced one > is definitely undefined code. Can you point to the section in the language ref for this? Pointer arithmetic? Sorry, I cannot provide the whole file; it's part of a commercial virtualization product. It's interfacing with (possibly proprietary virtual) hardware via a specification (which I have never seen). The subtraction of 0x2000 from the pointer is, apparently, to access specific hardware. If the operations are undefined, how should one interface directly with hardware without having code broken by UB?
[Bug c/119070] gcc15 incorrectly reporting negative array-bounds errors
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119070 --- Comment #1 from Taylor Hutt --- Created attachment 60616 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60616&action=edit C file that demonstrates errors.
[Bug tree-optimization/119070] gcc15 incorrectly reporting negative array-bounds errors
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119070
--- Comment #7 from Taylor Hutt ---
(In reply to Andrew Pinski from comment #6)
> You could do:
>
>struct_1 *v1 = &global_0.f_2_0;
>asm("":"+r"(v1));
>unsigned char *v2 = (unsigned char *)v1;
>
> to hide from GCC that the address of v2 is related to a global variable.
> And that should get rid of the warning too.
>
> But otherwise this is undefined code.
Why not cast the pointer to uintptr_t at the point of the undefined behavior
pointer arithmetic?
[Bug c/119070] gcc15 incorrectly reporting negative array-bounds errors
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119070 --- Comment #2 from Taylor Hutt --- Created attachment 60617 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60617&action=edit Shell script to compile C file.
[Bug c/119070] New: gcc15 incorrectly reporting negative array-bounds errors
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119070 Bug ID: 119070 Summary: gcc15 incorrectly reporting negative array-bounds errors Product: gcc Version: 15.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: c Assignee: unassigned at gcc dot gnu.org Reporter: taylor.hutt at broadcom dot com Target Milestone: --- The code in 'cit.c' has been extracted from a production audio driver present in widely used virtualization products. The code has not been meaningfully changed in many years. The gcc version banner is: x86_64-vmw-linux-gnu-gcc (crosstool-NG 1.26.0.127_1c395b2) 15.0.1 20250227 (experimental) Copyright (C) 2025 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. The code was checked out using crosstool-ng at the top of the 'master' branch on the day of the build. The script 'citi' (with modifications to produce a path to gcc15) will reproduce the errors. The errors are as follows: /tmp/gcc15/cit.c: In function 'function_1': /tmp/gcc15/cit.c:27:19: error: array subscript -227 is outside array bounds of 'struct_2[1]' [-Werror=array-bounds=] 27 |unsigned char *v3 = v2 - 0x2000; | ^~ /tmp/gcc15/cit.c:16:17: note: at offset -8192 into object 'global_0' of size 36 16 | extern struct_2 global_0; | ^~~~ /tmp/gcc15/cit.c:27:19: error: array subscript -227 is outside array bounds of 'struct_2[1]' [-Werror=array-bounds=] 27 |unsigned char *v3 = v2 - 0x2000; | ^~ /tmp/gcc15/cit.c:16:17: note: at offset -8192 into object 'global_0' of size 36 16 | extern struct_2 global_0; | ^~~~ /tmp/gcc15/cit.c:27:19: error: array subscript -227 is outside array bounds of 'struct_2[1]' [-Werror=array-bounds=] 27 |unsigned char *v3 = v2 - 0x2000; | ^~ /tmp/gcc15/cit.c:16:17: note: at offset -8192 into object 'global_0' of size 36 16 | extern struct_2 global_0; | ^~~~ cc1: all warnings being treated as errors I find, using https://godbolt.org/, that the regression was introduced in gcc10. The code compiles without issue with gcc9.
