[lldb-dev] Problem with dotest_channels.py

Todd Fiala via lldb-dev lldb-dev at lists.llvm.org
Mon Dec 14 14:03:22 PST 2015

Hi Zachary!

On Mon, Dec 14, 2015 at 1:28 PM, Zachary Turner via lldb-dev <
lldb-dev at lists.llvm.org> wrote:

> Hi Todd, lately I've been seeing this sporadically when running the test
> suite.
> [TestNamespaceLookup.py FAILED]
> Command invoked: C:\Python27_LLDB\x86\python_d.exe
> D:\src\llvm\tools\lldb\test\dotest.pyc -q --arch=i686 --executable
> D:/src/llvmbuild/ninja/bin/lldb.exe -s
> D:/src/llvmbuild/ninja/lldb-test-traces -u CXXFLAGS -u CFLAGS
> --enable-crash-dialog -C d:\src\llvmbuild\ninja_release\bin\clang.exe
> --results-port 55886 --inferior -p TestNamespaceLookup.py
> D:\src\llvm\tools\lldb\packages\Python\lldbsuite\test --event-add-entries
> worker_index=10:int
> 416 out of 416 test suites processed - TestAddDsymCommand.py
>     error: uncaptured python exception, closing channel
> <lldbsuite.test.dotest_channels.UnpicklingForwardingReaderChannel connected
> at 0x2bdd578> (<class 'socket.error'>:[Errno 10054] An
> existing connection was forcibly closed by the remote host
> [C:\Python27_LLDB\x86\lib\asyncore.py|read|83]
> [C:\Python27_LLDB\x86\lib\asyncore.py|handle_read_event|449]
> [D:\src\llvm\tools\lldb\packages\Python\lldbsuite\test\dotest_channels.py|handle_read|133]
> [C:\Python27_LLDB\x86\lib\asyncore.py|recv|387])
> It seems to happen randomly and not always on the same test.  Sometimes it
> doesn't happen at all.  I wonder if this could be related to some of the
> work that's been going on recently.  Are you seeing this?  Any idea how to
> diagnose?


That *looks* like one side of the connection between the inferior and the
test runner process choked on reading content from the test event socket
when the other end went down.  Reading it a bit more carefully, it looks
like it is the event collector (which would be the parallel test runner
side) that was receiving when the socket went down.

I think this means I just need to put a try block around the receiver and
just bail out gracefully (possibly with a message) when that happens at an
unexpected time.  Since test inferiors can die at any time, possibly due to
a timeout where they are forcibly killed, we do need to handle that

I'll see if I can force it, replicate it, and fix it.  I'll look at that
now (pending watching the buildbots for the other change I just put in).

And yes, this would be a code path that we use heavily with the xUnit
reporter, but only started getting used by you more recently when I turned
on the newer summary results by default.  (The newer summary results use
the test event system, which means test inferiors are now going to be using
the sockets to pass back test events, where you didn't have that happening
before unless you used the curses or xUnit results formatter).

I hope to have it reproduced and fixed up here quickly.  I suspect you may
have an environment that just might make it more prevalent, but it needs to
be fixed.

Hopefully back in a bit with a fix!

> _______________________________________________
> lldb-dev mailing list
> lldb-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20151214/1b9d8be0/attachment.html>

More information about the lldb-dev mailing list