Hello Guinevere,

Guinevere Larsen blarsen@redhat.com writes:

I just checked out this regression but I think there's something odd going on in the setup that is running the tests. When I run the test locally, the relevant part of the logs looks like this:

(gdb) PASS: gdb.threads/threadcrash.exp: test_live_inferior: continue to breakpoint: running to breakpoint     print $_inferior_thread_count^M     $1 = 7^M     (gdb) PASS: gdb.threads/threadcrash.exp: test_live_inferior: $thread_count == 7     thread apply all backtrace^M     ^M < lots of GDB output >     (gdb) PASS: gdb.threads/threadcrash.exp: test_live_inferior: Get thread information     PASS: gdb.threads/threadcrash.exp: test_live_inferior: $unwind_fail == false     PASS: gdb.threads/threadcrash.exp: test_live_inferior: $thread_count == [llength $test_list]

Whereas the failed log reads like this:

(gdb) PASS: gdb.threads/threadcrash.exp: test_live_inferior: continue to breakpoint: running to breakpoint     print $_inferior_thread_count     $1 = 7PASS: gdb.threads/threadcrash.exp: test_live_inferior: $thread_count == 7

(gdb) PASS: gdb.threads/threadcrash.exp: test_live_inferior: Get thread information     PASS: gdb.threads/threadcrash.exp: test_live_inferior: $unwind_fail == false     FAIL: gdb.threads/threadcrash.exp: test_live_inferior: $thread_count == [llength $test_list]     thread apply all backtrace

Notice how the gdb command to "apply all backtrace", which is used to create the test_list variable, happens after checking if the variable has been set correctly.

I have no clue how things can get so out of sync...

It gets out of sync because the gdb_test_multiple pattern in test_thread_count doesn't include the GDB prompt, so that function can return before GDB prints the prompt. In this case, this pattern in thread_apply_all will immediately match and make it return before constructing $test_list:

-re "$::gdb_prompt " { pass $gdb_test_name }

That is why the "Get thread information" test passed right after the GDB prompt printed after the "$thread_count == 7" test.

This patch makes the testcase work reliably for me, in a machine where I could often reproduce the failure from the CI:

diff --git a/gdb/testsuite/gdb.threads/threadcrash.exp b/gdb/testsuite/gdb.threads/threadcrash.exp index 3905ad6f9362..944fbcac1b18 100644 --- a/gdb/testsuite/gdb.threads/threadcrash.exp +++ b/gdb/testsuite/gdb.threads/threadcrash.exp @@ -28,7 +28,7 @@ proc test_thread_count {} { set thread_count 0

gdb_test_multiple "print $_inferior_thread_count" "getting thread count" { - -re ".* = ([0-9]+).*" { + -re ".* = ([0-9]+)\r\n$::gdb_prompt " { set thread_count $expect_out(1,string) gdb_assert {$thread_count == 7} }