parallel tests broken on Debian stable

Daniel Kahn Gillmor dkg at fifthhorseman.net
Mon May 20 17:44:20 PDT 2019


On Mon 2019-05-20 19:49:02 -0400, Daniel Kahn Gillmor wrote:
> However, i tried selectively upgrading all the versions of all of these
> packages *except for gdb* to the version in buster (or to the version
> from backports, in the case of the kernel).  and i'm *still* seeing the
> problem on the stretch system.
>
> So perhaps it's some interaction between timeout and gdb?  I haven't
> managed to test that particular combination yet.

Ok, i just went ahead and tested gdb 8.2.1-2 on a stretch system (after
rolling back my other package upgrades.

Getting gdb installed cleanly required upgrading libc6, libssl1.1, and
libpython (3.5 to 3.7) and associated packages to their versions from
buster, as well as pulling in a few libraries whose sonames weren't
already in stretch at all (libipt2, libncursesw6, libtinfo6).

once gdb was upgraded, the tests passed with moreutils parallel with
timeout enabled.

So the combination that tickles the problem is:

 * GNU timeout (either version)
 * moreutils parallel (either version)
 * gdb 7.12-6 (stretch)

and the problem goes away when gdb is upgraded to 8.2.1-2.

I wasn't able to whittle it down to a smaller reliable test,
unfortunately, but hopefully this is enough for someone else to pick it
up and run with it.


Digging around a bit further, i see back in 2011
9ade8160a6a3f6f55996d7a58dc2cd81a6df8395 from amdragon (cc'ed, though i
haven't heard from him since 2016) that introduced this comment (now in
tests/T380-atomicity.sh):

    # -tty /dev/null works around a conflict between the 'timeout' wrapper
    # and gdb's attempt to control the TTY.
     […]
    gdb -tty /dev/null -batch -x $TEST_DIRECTORY/atomicity.gdb notmuch >/dev/null 2>/dev/null


I tried adding "-tty /dev/null" to every place where ${TEST_GDB} was
invoked (in T050, T060, and T070), and i found that it produced test
failures in T050 and T060.  However, just directing stdin from /dev/null
on those runs of ${TEST_GDB} solved the problem for me. (see patch below)

Leaving any of those three gdb invocations without anything changed
meant that "make -j4 check" would hang pretty reliably with processes
stopped by SIGTTOU again with gdb from stretch.

Again, i don't understand any of this mechanism in detail, but the
following patch appears to avoid hangs of the test suite for me on a
stock debian stretch installation, and is (i think) a narrower change
than the earlier proposal that forced stdin from /dev/null on all the
tests uniformly.

If no one objects, i can propose this as an actual patch, but i'd feel
much happier if someone who understands this particular situation can
propose it with a clearer justifiation.

Regards,

        --dkg


diff --git a/test/T050-new.sh b/test/T050-new.sh
index dfc8508f..9b3d41df 100755
--- a/test/T050-new.sh
+++ b/test/T050-new.sh
@@ -366,7 +366,7 @@ run
 EOF
 
 ${TEST_GDB} --batch-silent --return-child-result -x notmuch-new-vanish.gdb \
-    --args notmuch new 2>OUTPUT 1>/dev/null
+    --args notmuch new 2>OUTPUT 1>/dev/null </dev/null
 echo "exit status: $?" >> OUTPUT
 
 # Clean up the file in case gdb isn't available.
diff --git a/test/T060-count.sh b/test/T060-count.sh
index 0c0bf473..a393edb6 100755
--- a/test/T060-count.sh
+++ b/test/T060-count.sh
@@ -116,7 +116,7 @@ EOF
 backup_database
 test_begin_subtest "error message from query_search_messages"
 ${TEST_GDB} --batch-silent --return-child-result -x count-files.gdb \
-    --args notmuch count --output=files '*' 2>OUTPUT 1>/dev/null
+    --args notmuch count --output=files '*' 2>OUTPUT 1>/dev/null < /dev/null
 cat <<EOF > EXPECTED
 notmuch count: A Xapian exception occurred
 A Xapian exception occurred performing query
diff --git a/test/T070-insert.sh b/test/T070-insert.sh
index 05be473a..2d281fe6 100755
--- a/test/T070-insert.sh
+++ b/test/T070-insert.sh
@@ -267,12 +267,14 @@ for code in  FILE_NOT_EMAIL READ_ONLY_DATABASE UPGRADE_REQUIRED PATH_ERROR; do
     test_begin_subtest "EXIT_FAILURE when index_file returns $code"
     test_expect_code 1 \
          "${TEST_GDB} --batch-silent --return-child-result \
+ 	     -tty /dev/null \
 	     -ex 'set args insert < $gen_msg_filename' \
 	     -x index-file-$code.gdb notmuch"
 
     test_begin_subtest "success exit with --keep when index_file returns $code"
     test_expect_code 0 \
          "${TEST_GDB} --batch-silent --return-child-result \
+ 	     -tty /dev/null \
 	     -ex 'set args insert --keep < $gen_msg_filename' \
 	     -x index-file-$code.gdb notmuch"
 done
@@ -281,12 +283,14 @@ for code in OUT_OF_MEMORY XAPIAN_EXCEPTION ; do
     test_begin_subtest "EX_TEMPFAIL when index_file returns $code"
     test_expect_code 75 \
          "${TEST_GDB} --batch-silent --return-child-result \
+ 	     -tty /dev/null \
 	     -ex 'set args insert < $gen_msg_filename' \
 	     -x index-file-$code.gdb notmuch"
 
     test_begin_subtest "success exit with --keep when index_file returns $code"
     test_expect_code 0 \
          "${TEST_GDB} --batch-silent --return-child-result \
+ 	     -tty /dev/null \
 	     -ex 'set args insert --keep < $gen_msg_filename' \
 	     -x index-file-$code.gdb notmuch"
 done
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 227 bytes
Desc: not available
URL: <http://notmuchmail.org/pipermail/notmuch/attachments/20190520/a107c14d/attachment-0001.sig>


More information about the notmuch mailing list