]> git.ipfire.org Git - thirdparty/gcc.git/commit
strlen: Fix up !si->full_string_p handling in count_nonzero_bytes_addr [PR115152]
authorJakub Jelinek <jakub@redhat.com>
Wed, 22 May 2024 07:13:50 +0000 (09:13 +0200)
committerJakub Jelinek <jakub@redhat.com>
Wed, 22 May 2024 07:13:50 +0000 (09:13 +0200)
commitdbc9b45a3c2468ad134b3a9bd4961f7ae6bc1e67
treeb0938a3008b5ae6d1254a5938e7cf1fa3dbf0662
parentd3c506eff54fcbac389a529c2e98da108a410b7f
strlen: Fix up !si->full_string_p handling in count_nonzero_bytes_addr [PR115152]

The following testcase is miscompiled because
strlen_pass::count_nonzero_bytes_addr doesn't handle correctly
the !si->full_string_p case.
If si->full_string_p, it correctly computes minlen and maxlen as
minimum and maximum length of the '\0' terminated stgring and
clears *nulterm (ie. makes sure !full_string_p in the ultimate
caller) if minlen is equal or larger than nbytes and so
'\0' isn't guaranteed to be among those bytes.
But in the !si->full_string_p case, all we know is that there
are [minlen,maxlen] non-zero bytes followed by unknown bytes,
so effectively the maxlen is infinite (but caller cares about only
the first nbytes bytes) and furthermore, we never know if there is
any '\0' char among those, so *nulterm needs to be always cleared.

2024-05-22  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/115152
* tree-ssa-strlen.cc (strlen_pass::count_nonzero_bytes_addr): If
!si->full_string_p, clear *nulterm and set maxlen to nbytes.

* gcc.dg/pr115152.c: New test.
gcc/testsuite/gcc.dg/pr115152.c [new file with mode: 0644]
gcc/tree-ssa-strlen.cc