Hi, On 2025-08-27 23:25, Bo YU wrote: > On Wed, Aug 27, 2025 at 11:24 AM Tianon Gravi <[email protected]> wrote: > > > > On Tue, 26 Aug 2025 at 20:09, Bo YU <[email protected]> wrote: > > > Now golang-1.24 has one FTBFS on riscv64 due to: > > > > > > ``` > [...] > > > memory mapping is not match. Maybe we need to skip it explicitly? > > > > > > I will test it with skipping the test case. > > > > > > [0]: https://github.com/golang/go/issues/64345 > > > > I did some testing of this and found that it seems to work fine on a > > completely emulated riscv64 VM in QEMU, but on the real hardware that > > is the ricci porterbox I often got even worse than the buildds and hit > > segfaults instead that looked like the one in > > https://github.com/golang/go/issues/57691#issuecomment-2962825755 > > > > yeah, qemu riscv64 on my system is sv57 and Unmatched(porterbox and > buildd machines) > is sv39. > > > I was trying to figure out whether my -3 upload is what broke this, > > but I was ultimately unable to reproduce the same failure the buildds > > are having on either -3 *or* -1. For what it's worth, both gcc-12 > > version 12.4.0-8 and 12.5.0-1 were uploaded since the last successful > > buildd build of src:golang-1.24, so it's possible one of those is > > related too (the source of libtsan and the error message we're seeing > > in the tests). > > > > indeed. Very high probability is libtsan2_15.2.0-1 maybe...
Yep, I confirm that. The issue is reproducible in a trixie environment with only gcc-15-base and libtsan2 from forky. Building golang-1.24 with gcc-15 doesn't fix the issue, so it's not due to the version mismatch between the compiler and library. It rather seems to be a new feature or bug in libtsan2. Regards Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B [email protected] http://aurel32.net
