[PATCH v10] Add pretty printers for the NPTL lock types

[Stefan Liebler]

On 07/18/2016 08:50 PM, Martin Galvan wrote:
> Changes from v9:

> Besides that, I've added a few more checks to the test harness. We now check for a minimum Pexpect
> version (4) and a minimum gdb version (7.7), as well as if gdb actually has Python support. I also
> discarded the custom gdb prompt since we're not reading a .gdbinit anymore. Notice gdb 7.7 still
> has the RegexpCollectionPrettyPrinter bug fixed for 7.8; if you're using a gdb older than 7.8 you
> should fix it before running the tests (the README points out the gdb commit that fixes it).
As a note: gdb is installed on all my machines in a version > 7.8. Thus 
I could not test this checks.

I run the tests on a machine without pexpect => UNSUPPORTED:
The out-files contain the hint of missing pexpect. Ok.

I run the tests on a machine with pexpect version 3.1 => UNSUPPORTED:
The out-files contain the hint of missing the correct pexpect version. Ok.

I run the tests on a machine (fedora23 i686, s390x, s390) with pexpect 
version 4.0.1 => FAIL:
The out-file is empty, but on terminal the tests dump:
Traceback (most recent call last):
   File "test-mutexattr-printers.py", line 22, in <module>
     from test_printers_common import *
   File "/home/stli/glibcDir/glibc/scripts/test_printers_common.py", 
line 67, in <module>
     gdb_version = (int(match.group(1)), int(match.group(2)))
AttributeError: 'NoneType' object has no attribute 'group'

re.search  does not find the gdb-version string and is returning None.
On e.g. Fedora, gdb --version prints:
GNU gdb (GDB) Fedora 7.10-23.fc23
Copyright (C) 2015 Free Software Foundation, Inc.
...


> As for the issues Stefan was seeing, I've tried this myself using gdb 7.10 and couldn't reproduce
> them. It seems that in your case gdb.lookup_type isn't finding the struct types representing the attributes.
> One thing that might be causing the issue is your libpthread.so not being compiled with debugging symbols.
>  From what I saw, the DWARF DIEs describing the attr structs come from libpthread, not the test programs.
> Could you check if your libpthread is stripped, and if so re-compile it with debugging symbols?
>
If I remove the check for correct gdb-version as these machines (i686, 
s390x, s390 fedora 23) have correct gdb, pexpect version and glibc is 
built without debug-info:
FAIL: nptl/test-condattr-printers; out-file:
Error: Response does not match the expected pattern.
Command: print *attr
Expected pattern: pthread_condattr_t
Response: Python Exception <class 'gdb.error'> No struct type named 
pthread_condattr.:
Python Exception <class 'gdb.error'> No struct type named pthread_condattr.:
$3 = {__size = "\000\000\000", __align = 0}
(gdb)

FAIL: nptl/test-mutexattr-printers
FAIL: nptl/test-rwlockattr-printers



If glibc is built on i686 fedora 23 with debug-info:
FAIL: nptl/test-condattr-printers; out-file:
Error: Response does not match the expected pattern.
Command: print *condvar
Expected pattern: Shared = Yes
Response: $6 = pthread_cond_t = {Threads waiting for this condvar = 0,
Clock ID = CLOCK_REALTIME, Shared = No, Mutex = Cannot access memory at 
address 0xffffffff
(gdb)


If glibc is built on s390x (64bit) fedora 23 with debug-info:
All nptl pretty-printer tests succeed.

If glibc is built on s390 (31bit) fedora 23 with debug-info:
FAIL: nptl/test-condattr-printers;
The out-file contains the same error as on 32bit intel.




Is this the intention, that glibc has to be build with debug-information 
in order to get no test fails?

Bye
Stefan