https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126952
Zhongteng Gui <dragon-archer at outlook dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |dragon-archer at outlook dot
com
--- Comment #10 from Zhongteng Gui <dragon-archer at outlook dot com> ---
Hi!
I observed the same bootstrap failure on MSYS2 while building GCC 17-20261011
with the MSYS2 GCC 15.3.0 toolchain.
Configuration:
target: x86_64-pc-cygwin
GCC snapshot: 17-20261011
source commit: b6822b2c3b6c2564600f90b9115421f96fc6fb0f
configure: --enable-bootstrap --enable-checking=release
stage2 flags: -g -O2 -fno-checking -gtoggle
Original CI job:
https://github.com/dragon-archer/gcc-for-windows/actions/runs/38732712818/job/116414720915#step:5:10579
gcc/wide-int.h:1272:73: error:
'...wide_int_storage::u...valp' may be used uninitialized
[-Werror=maybe-uninitialized]
gcc/gimple-ssa-warn-access.cc:2870:21: note: 'rng' declared here
cc1plus: all warnings being treated as errors
The inlining path is:
wide_int_storage::write_val -> wi::copy
-> wide_int_storage::operator= -> generic_wide_int::operator=
-> irange_bitmask::set_unknown -> irange::set_varying
-> int_range<N, RESIZABLE>::int_range(tree) -> memmodel_to_uhwi
This matches the bisected revision mentioned here:
r17-3330-g4fe2072d5d16ce
tree-optimization/126887 - allow backward threading to loop exit
The proposed constructor initialization of `irange::m_bitmask` with the type
precision looks applicable to this failure. I have started a new CI test with
the proposed patch from
https://gcc.gnu.org/pipermail/gcc-patches/2026-October/733683.html at
https://github.com/dragon-archer/gcc-for-windows/actions/runs/38780295782/job/116579984012
As of writing, that job is still running.
