[#85349] [Ruby trunk Bug#14334] Segmentation fault after running rspec (ruby/2.5.0/erb.rb:885 / simplecov/source_file.rb:85) — pragtob@...
Issue #14334 has been updated by PragTob (Tobias Pfeiffer).
3 messages
2018/02/02
[#85358] Re: [ruby-cvs:69220] nobu:r62039 (trunk): compile.c: unnecessary freezing — Eric Wong <normalperson@...>
nobu@ruby-lang.org wrote:
5 messages
2018/02/03
[#85612] Why require autoconf 2.67+ — leam hall <leamhall@...>
Please pardon the intrusion; I am new to Ruby and like to pull the
6 messages
2018/02/17
[#85634] [Ruby trunk Bug#14494] [PATCH] tool/m4/ruby_replace_type.m4 use AC_CHECK_TYPES for HAVE_* macros — normalperson@...
Issue #14494 has been reported by normalperson (Eric Wong).
3 messages
2018/02/19
[#85674] [Ruby trunk Feature#13618] [PATCH] auto fiber schedule for rb_wait_for_single_fd and rb_waitpid — matz@...
Issue #13618 has been updated by matz (Yukihiro Matsumoto).
5 messages
2018/02/20
[#85686] Re: [Ruby trunk Feature#13618] [PATCH] auto fiber schedule for rb_wait_for_single_fd and rb_waitpid
— Eric Wong <normalperson@...>
2018/02/20
matz@ruby-lang.org wrote:
[#85704] Re: [Ruby trunk Feature#13618] [PATCH] auto fiber schedule for rb_wait_for_single_fd and rb_waitpid
— Koichi Sasada <ko1@...>
2018/02/21
On 2018/02/20 18:06, Eric Wong wrote:
[ruby-core:85462] Re: [Ruby trunk Bug#14357] thread_safe tests suite segfaults
From:
Vladimir Makarov <vmakarov@...>
Date:
2018-02-07 15:36:32 UTC
List:
ruby-core #85462
On 02/06/2018 02:38 PM, Eric Wong wrote: > Vladimir Makarov <vmakarov@redhat.com> wrote: >> On 02/06/2018 05:00 AM, Eric Wong wrote: >>> during rebuild. Disabling the free(tab->entries) at line >>> st.c:792 (patch below) seems to indicate success with the >>> thread_safe test suite (letting it loop overnight). > It still crashed after four runs :< It might run longer with > the simplecov/coveralls stuff commented out in spec_helper.rb > since coverage creates a giant hash and might increase the chance > of failure. > >> Eric, thank you for working on the problem and analyzing it. I'll look at >> this and try to fix it as soon as possible. > I reproduced this crash although the reproducing is not stable with or without valgrind. It is a typical data race. The same problem existed in the **old hash tables**. It also rebuilt tables and freed old data structure. **File st.c was never thread-safe**. The data races are/were possible in many places. We could make st.c thread-safe. But I don't think it is a right way. It is not a trivial task and it also will hurt performance considerably. We still needs thread-unaware level to work with hash tables (st.c) for cases when tables are used internally in one thread. So I think the crash should be fixed in other places where calls of st.c happen. I don't know how it should be fixed because I don't know Ruby thread semantics. Does Ruby guarantee that there are no data races or should a ruby programmer still provides thread synchronization despite GIL? If it is later, thread_safe gem is probably buggy because one thread reading a table and another thread inserting elements while process table in a Ruby block. If there is no sync it is a typical data race and the result is unpredictable. In this case it is a segfault crash. We could just give a better message about the data races if segfault happens in st.c. Also I don't know how GIL works. Where the thread switching can happen. Is the switch possible in find_table_ind or we just read unsync cashed value in the thread because st.c never used atomics. Unfortunately I am not well familiar with Ruby threads so it is hard for me to say how to fix it. I only think that we should keep st.c thread-unaware as it always was. Unsubscribe: <mailto:ruby-core-request@ruby-lang.org?subject=unsubscribe> <http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>