Re: [Discuss-gnuradio] Debugging polar code crashes

[Cinaed Simson]

On 07/17/2017 03:20 AM, Marcus Müller wrote:
> Hi Johannes,
> 
> thanks for this very short and informal bug report on your own software :)
> 
> so, a bit of debugging from my side:
> 
> gdb --args python2 $(which gnuradio-companion)
> (gdb) run

Cool! That's how you debug the XML code. I was wondering how to do that
last week.

-- Cinaed

> 
> …
> 
> Trigger hangup by dragging POLAR config onto canvas
> 
> …
> 
> [ctrl-C]
> Thread 1 "python2" received signal SIGINT, Interrupt.
> (gdb) backtrace
> #0  0x00007ffff7a76fbe in instancemethod_call.lto_priv () at 
> /lib64/libpython2.7.so.1.0
> #1  0x00007ffff7a2bea3 in PyObject_Call () at /lib64/libpython2.7.so.1.0
> #2  0x00007ffff7a7d323 in slot_sq_item.lto_priv () at 
> /lib64/libpython2.7.so.1.0
> #3  0x00007ffff7a47b8f in iter_iternext.lto_priv () at 
> /lib64/libpython2.7.so.1.0
> #4  0x00007ffff7a2d2de in PyIter_Next () at /lib64/libpython2.7.so.1.0
> …
> 
> aha, so that's happening in python-land. Thus, throw python debugging at
> this:
> 
> (gdb) py-bt
> Traceback (most recent call first):
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/core/Param.py",
>  line 136, in __getitem__
> *return str(self._param.get_opt(item)) if self._param.is_enum() else
> NotImplemented*
>   File "cheetah_DynamicallyCompiledCheetahTemplate_1500284883_72_24761.py", 
> line 85, in respond
>   File "/usr/lib64/python2.7/site-packages/Cheetah/Template.py", line 1005, 
> in __str__
>     rc = getattr(self, mainMethName)()
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/core/Block.py",
>  line 703, in resolve_dependencies
>     return str(Template(tmpl, n))
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/core/Param.py",
>  line 338, in get_hide
>     hide = self.get_parent().resolve_dependencies(self._hide).strip()
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/Block.py",
>  line 211, in create_labels
>     markups = [param.get_markup() for param in self.get_params() if 
> param.get_hide() not in ('all', 'part')]
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/Element.py",
>  line 78, in create_labels
>     for child in self.get_children():child.create_labels()
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/FlowGraph.py",
>  line 476, in update
>     self.create_labels()
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/ActionHandler.py",
>  line 113, in flow_graph_update
>     fg.update()
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/ActionHandler.py",
>  line 358, in _handle_action
>     flow_graph_update()
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/Actions.py",
>  line 114, in __call__
>     self.emit('activate')
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/FlowGraph.py",
>  line 160, in add_new_block
>     Actions.ELEMENT_CREATE()
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/gui/DrawingArea.py",
>  line 96, in _handle_drag_data_received
>     self._flow_graph.add_new_block(selection_data.data, (x, y))
>   File 
> "/home/marcus/.usrlocal/lib64/python2.7/site-packages/gnuradio/grc/main.py", 
> line 54, in main
>     gtk.main()
>   File "/home/marcus/.usrlocal/bin/gnuradio-companion", line 92, in run_main
>     exit(main())
>   File "/home/marcus/.usrlocal/bin/gnuradio-companion", line 99, in <module>
>     run_main()
> 
> *Emphasis* by me. OK, so it's a ["string"] dereferral problem in Cheetah
> calling back into GRC code, most likely (the point in the execution time
> where my ctrl-C hits the python process is of course a random variable,
> with the event-space being all lines of code executed in that infinite
> loop in a "this eats up CPU" situation; but there's an accumulation
> point at the things that are called most often, so heh, this "works on
> average", or after a few tries. This time, it worked on the first
> ctrl-C). So, Cheetah looping over something it parses. Ok, we have a
> suspect. Let's look at the block definition XML file:
> 
> <?xml version="1.0"?>
> <block>
>   <name>POLAR code Configurator</name>
>   <key>variable_polar_code_configurator</key>
>   <import>from gnuradio.fec import polar</import>
>   <var_make>self.$(id) = $(id) = polar.load_frozen_bits_info(False, $channel, 
> $block_size, $num_info_bits, $design_snr, $mu)</var_make>
>   <var_value>polar.load_frozen_bits_info(True, $channel, $block_size, 
> $num_info_bits, $design_snr, $mu)</var_value>
>   <make></make>
> 
>   <param>
>     <name>Channel</name>
>     <key>channel</key>
>     <value>polar.CHANNEL_TYPE_BEC</value>
>     <type>string</type>
>       <option>
>         <name>BEC</name>
>         <key>polar.CHANNEL_TYPE_BEC</key>
>       </option>
>       <option>
>         <name>AWGN</name>
>         <key>polar.CHANNEL_TYPE_AWGN</key>
>       </option>
>   </param>
> 
>   <param>
>     <name>Block size (N)</name>
>     <key>block_size</key>
>     <type>int</type>
>   </param>
> 
>   <param>
> *<name>#Info Bits (K)</name>*
>     <key>num_info_bits</key>
>     <type>int</type>
>   </param>
> 
>   <param>
>     <name>design SNR</name>
>     <key>design_snr</key>
>     <value>0.0</value>
>     <type>float</type>
>   </param>
> 
>   <param>
>     <name>mu</name>
>     <key>mu</key>
>     <value>16</value>
>     <type>int</type>
> *<hide>#if 'BEC' in $getVar('channel') then 'all' else 'none' #</hide>*
>   </param>
> </block>
> 
> Hm, that <hide> element looks fishy.
> Removed it, tried again, works.
> 
> So, you've got a bug in the way your <hide> element is parsed. Maybe
> it's a combination with the (probably cheetah-illegal?) # at the
> beginning of "num_info_bits"'s <name> element. Hope that gives you a
> direction to look at!
> 
> By the way, wouldn't we use small k instead of K and n instead of N for
> a (n,k) code?
> 
> Cheers,
> Marcus
> 
> On 07/17/2017 11:03 AM, Johannes Demel wrote:
>> Hi List,
>>
>> I created an issue for this on github [0]. I assume the bug manifested
>> itself due to a change in GRC.
>>
>> Cheers
>> Johannes
>>
>>
>> [0] https://github.com/gnuradio/gnuradio/issues/1385
>>
>>
>> On 15.07.2017 04:36, Cinaed Simson wrote:
>>> Hi Johannes - I captured the output of grcc for both polar grc files and
>>> they can't find
>>>
>>>    polar_config
>>>
>>> I did a search of the gnuradio tree and my home directory
>>>
>>>    find . -name polar_config\*
>>>
>>> and I couldn't find it either.
>>>
>>> -- Cinaed
>>>
>>>
>>> On 07/14/2017 01:04 AM, Johannes Demel wrote:
>>>> Hi,
>>>>
>>>> this is a tricky one. My assumption is, that the configurator block
>>>> tries to compute channel capacities with a complex algorithm. There is
>>>> also the BEC channel approximation, which should be called.
>>>> This whole bug is caused by some GRC changes, as far as I could figure
>>>> it out. Could you try to have a look into the python code of the block
>>>> configurator and make sure, only the BEC functions are called? This
>>>> should fix your GRC problem.
>>>>
>>>> The configurator automates quite a few things regarding channel
>>>> construction and has a built-in system for cached channel capacities. I
>>>> assume something does not behave as expected here.
>>>>
>>>> Cheers
>>>> Johannes
>>>>
>>>> On 14.07.2017 01:47, Cinaed Simson wrote:
>>>>> On 07/13/2017 12:03 AM, Johannes Demel wrote:
>>>>>> Hi Alex,
>>>>>>
>>>>>> could you be more specific about
>>>>>> 'When I drop the POLAR code configurator block on the canvas, the
>>>>>> gnuradio program turns down.'?
>>>>>> Does GRC quit with an error? Does it turn dark and is
>>>>>> unresponsive? Is
>>>>>> there anything printed on the commandline?
>>>>>>
>>>>>> Cheers
>>>>>> Johannes
>>>>>
>>>>> Hey, I had that problem too!
>>>>>
>>>>> In my case, all the tests pass - I have the correct version of GSL
>>>>> installed - namely 1.16
>>>>>
>>>>> But I couldn't even load the polar grc files.
>>>>>
>>>>> When I press open on the grc it hangs - the grc was consuming 100% of
>>>>> the cpu.
>>>>>
>>>>> Eventually I had to kill it.
>>>>>
>>>>> I first noticed it in gnuradio-3.7.10. I'm currently running
>>>>> gnuradio-3.7.12git and I just checked and the problem is still there.
>>>>>
>>>>> And I can't generate the python code using grcc - both polar grc
>>>>> examples generate errors.
>>>>>
>>>>> -- Cinaed
>>>>>
>>>>>
>>>>>>
>>>>>>
>>>>>> On 12.07.2017 18:04, Marcus Müller wrote:
>>>>>>> Those test failures aren't great, but I think they might be bugs
>>>>>>> in the
>>>>>>> tests.
>>>>>>>
>>>>>>> So, OK, can't remember what you need to do under Ubuntu to get
>>>>>>> the full
>>>>>>> Python debugging capability in GDB, but I wrote a (slightly dated,
>>>>>>> since
>>>>>>> the python commands have improved) tutorial on debugging crashing
>>>>>>> Python
>>>>>>> applications [1].
>>>>>>> The idea is to simply execute your minimal crashing example using
>>>>>>>
>>>>>>> gdb --args python2 /path/to/flowgraph.py
>>>>>>> (gdb) run
>>>>>>> <wait for crash>
>>>>>>> (gdb) backtrace
>>>>>>>
>>>>>>> and figure out why where how things crash. If you got a
>>>>>>> backtrace, and
>>>>>>> don't immediately see how to proceed (hint: print a few local
>>>>>>> variables,
>>>>>>> if possible, using gdb's "print" command), share it with the mailing
>>>>>>> list
>>>>>>>
>>>>>>> Best regards,
>>>>>>> Marcus
>>>>>>>
>>>>>>> [1]https://wiki.gnuradio.org/index.php/TutorialsGDB
>>>>>>> On 07/12/2017 05:50 PM, Alex Homero Rivadeneira Erazo wrote:
>>>>>>>> I am using ubuntu 16.04 LTS.
>>>>>>>>
>>>>>>>> At the end of the make test, I have obtained some failures:
>>>>>>>>
>>>>>>>> 97% tests passed, 6 tests failed out of 218
>>>>>>>>
>>>>>>>> Total Test time (real) = 142.49 sec
>>>>>>>>
>>>>>>>> The following tests FAILED:
>>>>>>>> 48 - qa_ctrlport_probes (Failed)
>>>>>>>> 59 - qa_cpp_py_binding (Failed)
>>>>>>>> 76 - qa_cpp_py_binding_set (Failed)
>>>>>>>> 124 - qa_filterbank (Failed)
>>>>>>>> 154 - qa_ofdm_txrx (Failed)
>>>>>>>> 197 - qa_fading_model (Failed)
>>>>>>>> Errors while running CTest
>>>>>>>> Makefile:61: recipe for target 'test' failed
>>>>>>>> make: *** [test] Error 8
>>>>>>>>
>>>>>>>> Best regards,
>>>>>>>>
>>>>>>>> Alex
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On Wed, Jul 12, 2017 at 11:46 AM, Marcus Müller
>>>>>>>> <address@hidden <mailto:address@hidden>> wrote:
>>>>>>>>
>>>>>>>>     That might still give us enough debug info. Which Linux
>>>>>>>> distro are
>>>>>>>>     we talking about (or, is it Linux at all?)
>>>>>>>>
>>>>>>>>     Best regards,
>>>>>>>>
>>>>>>>>     Marcus
>>>>>>>>
>>>>>>>>
>>>>>>>>     On 07/12/2017 05:30 PM, Alex Homero Rivadeneira Erazo wrote:
>>>>>>>>>     Hi Marcus
>>>>>>>>>
>>>>>>>>>     At this moment, I am installing the
>>>>>>>>>     gnuradio 3.7.12git-171-get72c77bc from source, which is the
>>>>>>>>> same
>>>>>>>>>     that I installed with pybombs . Unfortunately, I just already
>>>>>>>>>     built the gnuradio with "make && make test". The gnuradio
>>>>>>>>> that I
>>>>>>>>>     had with the standard installation was gnuradio 3.7.9
>>>>>>>>>
>>>>>>>>>     Thanks
>>>>>>>>>
>>>>>>>>>     Best regards,
>>>>>>>>>
>>>>>>>>>     Alex
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>     On Wed, Jul 12, 2017 at 11:13 AM, Marcus Müller
>>>>>>>>>     <address@hidden <mailto:address@hidden>>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>         Hi Alex,
>>>>>>>>>
>>>>>>>>>         that does indeed sound like a bug. Things shouldn't be
>>>>>>>>> crashing!
>>>>>>>>>
>>>>>>>>>         So, what's the distribution you're working with? Maybe
>>>>>>>>> I can
>>>>>>>>>         help you a bit with the debugging. If you're, at the
>>>>>>>>> moment,
>>>>>>>>>         building GNU Radio, make sure that the CMake parameters
>>>>>>>>>         contain -DCMAKE_BUILD_TYPE=RelWithDebInfo , so that we get
>>>>>>>>>         debugging symbols :)
>>>>>>>>>
>>>>>>>>>         Best regards,
>>>>>>>>>
>>>>>>>>>         Marcus
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>         On 07/12/2017 05:11 PM, Alex Homero Rivadeneira Erazo
>>>>>>>>> wrote:
>>>>>>>>>>         Hi Marcus
>>>>>>>>>>
>>>>>>>>>>         Based on your suggestion I have deleted my Nabbles
>>>>>>>>>> account.
>>>>>>>>>>
>>>>>>>>>>         Respect to my problem with the POLAR code
>>>>>>>>>>         configurator block, I checked the discussion threads, but
>>>>>>>>>>         anyone gave a solution.
>>>>>>>>>>
>>>>>>>>>>         Note that this problem does not happen with
>>>>>>>>>>         a gnuradio installed with the distribution's package
>>>>>>>>>> manager
>>>>>>>>>>         (standard installation). However, the problem with this
>>>>>>>>>>         installation method is that gnuradio is not up to date
>>>>>>>>>> and
>>>>>>>>>>         it does not have all the blocks that I need in my
>>>>>>>>>> work. That
>>>>>>>>>>         is why I used pybombs to install gnuradio which has the
>>>>>>>>>>         blocks that I need. Now the problem is that gnuradio with
>>>>>>>>>>         pybombs does work with the POLAR code configurator block.
>>>>>>>>>>         When I drop the POLAR code configurator block on the
>>>>>>>>>> canvas,
>>>>>>>>>>         the gnuradio program turns down.
>>>>>>>>>>
>>>>>>>>>>         At this moment, I am installing gnuradio from source.
>>>>>>>>>> If you
>>>>>>>>>>         have other ideas please tell me.
>>>>>>>>>>
>>>>>>>>>>         Thanks
>>>>>>>>>>
>>>>>>>>>>         Best regards,
>>>>>>>>>>
>>>>>>>>>>         Alex
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>         On Wed, Jul 12, 2017 at 9:43 AM, Marcus Müller
>>>>>>>>>>         <address@hidden
>>>>>>>>>> <mailto:address@hidden>>
>>>>>>>>>>         wrote:
>>>>>>>>>>
>>>>>>>>>>             Hi Alex,
>>>>>>>>>>
>>>>>>>>>>             I hope you do get an answer, but that mail from
>>>>>>>>>> Jose is
>>>>>>>>>>             from the
>>>>>>>>>>             beginning of 2016 – not extremely likely.
>>>>>>>>>>
>>>>>>>>>>             Anyway, have you seen the other answers that Jose
>>>>>>>>>> got?
>>>>>>>>>>             If not: That's Nabble's fault. I never tire to
>>>>>>>>>> repeat:
>>>>>>>>>>             Nabble is just a terrible, terrible web frontend
>>>>>>>>>> to use
>>>>>>>>>>             the official GNU
>>>>>>>>>>             Radio Mailing list.
>>>>>>>>>>
>>>>>>>>>>             It's not the official mailing list archive. It never
>>>>>>>>>>             was, and it never
>>>>>>>>>>             will be. The official way to sign up for the mailing
>>>>>>>>>> list is
>>>>>>>>>>
>>>>>>>>>>            
>>>>>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>>>>>>>>
>>>>>>>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio>
>>>>>>>>>>
>>>>>>>>>>             and there's a link to the official mailing list
>>>>>>>>>> archive,
>>>>>>>>>>             too.
>>>>>>>>>>
>>>>>>>>>>             I'd recommend deleting your Nabble account, and
>>>>>>>>>> directly
>>>>>>>>>>             signing up
>>>>>>>>>>             under the above address with your email account. It's
>>>>>>>>>>             really easier to
>>>>>>>>>>             use a proper email client (eg. GMail's web
>>>>>>>>>> frontend) and
>>>>>>>>>>             simply the
>>>>>>>>>>             mails than Nabble. Nabble breaks discussion
>>>>>>>>>> threads, it
>>>>>>>>>>             makes answers
>>>>>>>>>>             invisible, and it has cost us hours and hours of
>>>>>>>>>>             unnecessary work so
>>>>>>>>>>             far. PLEASE, don't use Nabble.
>>>>>>>>>>
>>>>>>>>>>             Best regards,
>>>>>>>>>>
>>>>>>>>>>             Marcus
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>             On 07/12/2017 01:52 PM, Alex Rivadeneira wrote:
>>>>>>>>>>             > Dear Jose
>>>>>>>>>>             >
>>>>>>>>>>             > Can you tell me if you solved the problem with the
>>>>>>>>>>             POLAR code configurator
>>>>>>>>>>             > block? because I have the same problem in the
>>>>>>>>>> gnuradio
>>>>>>>>>>             installed with
>>>>>>>>>>             > pybombs.
>>>>>>>>>>             >
>>>>>>>>>>             > Thanks
>>>>>>>>>>             >
>>>>>>>>>>             > Alex
>>>>>>>>>>             >
>>>>>>>>>>             >
>>>>>>>>>>             >
>>>>>>>>>>             > --
>>>>>>>>>>             > View this message in context:
>>>>>>>>>>
>>>>>>>>>> http://gnuradio.4.n7.nabble.com/polar-code-example-tp58407p64520.html
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> <http://gnuradio.4.n7.nabble.com/polar-code-example-tp58407p64520.html>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>             > Sent from the GnuRadio mailing list archive at
>>>>>>>>>> Nabble.com.
>>>>>>>>>>             >
>>>>>>>>>>             > _______________________________________________
>>>>>>>>>>             > Discuss-gnuradio mailing list
>>>>>>>>>>             > address@hidden
>>>>>>>>>> <mailto:address@hidden>
>>>>>>>>>>             >
>>>>>>>>>>            
>>>>>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>>>>>>>>
>>>>>>>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>             _______________________________________________
>>>>>>>>>>             Discuss-gnuradio mailing list
>>>>>>>>>>             address@hidden
>>>>>>>>>> <mailto:address@hidden>
>>>>>>>>>>            
>>>>>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>>>>>>>>
>>>>>>>>>> <https://lists.gnu.org/mailman/listinfo/discuss-gnuradio>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>         --
>>>>>>>>>>         Regards,
>>>>>>>>>>
>>>>>>>>>>         Alex Homero Rivadeneira E.
>>>>>>>>>>         Cell: 438 397-8330 <tel:%28438%29%20397-8330>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>     --
>>>>>>>>>     Regards,
>>>>>>>>>
>>>>>>>>>     Alex Homero Rivadeneira E.
>>>>>>>>>     Cell: 438 397-8330 <tel:%28438%29%20397-8330>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> -- 
>>>>>>>> Regards,
>>>>>>>>
>>>>>>>> Alex Homero Rivadeneira E.
>>>>>>>> Cell: 438 397-8330
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Discuss-gnuradio mailing list
>>>>>>> address@hidden
>>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Discuss-gnuradio mailing list
>>>>>> address@hidden
>>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>>>> .
>>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Discuss-gnuradio mailing list
>>>>> address@hidden
>>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>>>
>>>>
>>>> _______________________________________________
>>>> Discuss-gnuradio mailing list
>>>> address@hidden
>>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>>
>>>
>>>
>>> _______________________________________________
>>> Discuss-gnuradio mailing list
>>> address@hidden
>>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>>>
>>
>> _______________________________________________
>> Discuss-gnuradio mailing list
>> address@hidden
>> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
> 
> 
> 
> _______________________________________________
> Discuss-gnuradio mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>