On Tue, Jun 5, 2012 at 9:20 AM, Jakub Jelinek <ja...@redhat.com> wrote: > On Mon, Jun 04, 2012 at 11:19:46PM -0700, Ian Lance Taylor wrote: >> This patch to libgo includes the TLS size in the requested stack size of >> a new thread, if possible. This relies on the glibc-specific (and >> undocumented) _dl_get_tls_static_info call. This is particularly >> necessary when using glibc, because glibc removes the static TLS size >> from the requested stack space, and gives an error if there is not >> enough space. That means that a program that has a lot of TLS variables >> will fail bizarrely, or worse may simply get a stack overflow >> segmentation violation at runtime. This patch is far from perfect, but >> at least works around that problem for such programs. Bootstrapped and >> ran Go testsuite on x86_64-unknown-linux-gnu. Committed to mainline and >> 4.7 branch. > > That is a very bad idea. _dl_get_tls_static_info is @@GLIBC_PRIVATE symbol, > therefore it must not be used by anything but glibc itself, may change or > may be removed at any time. Not using of GLIBC_PRIVATE symbols are even > enforced > by rpm, so if you build gcc as rpm, it won't install. So especially > committing that to 4.7 branch is fatal. > > Talk to libc-al...@sourceware.org for what should be done instead.
Ian, can you please revert the patch ASAP as I want to get 4.7.1 out of the door (well, a release candidate)? Otherwise we'll ship 4.7.1 with broken Go (not that we technically care - Go is not a primary language). Thanks, Richard. > Jakub
