[PATCH v2] tester: Add script to generate html coverage report from covoar output

Joel Sherrill joel at rtems.org
Mon Jun 4 19:21:46 UTC 2018 

On Mon, Jun 4, 2018 at 2:12 PM, Vijay Kumar Banerjee <
vijaykumar9597 at gmail.com> wrote:

> On 5 June 2018 at 00:31, Joel Sherrill <joel at rtems.org> wrote:
>
>> I will add that covoar was originally designed to generate a report on one
>> set at a time. The iteration over symbol sets was done at the scripting
>> level above that.
>>
>> This had the advantage of being simple at the time. It may still be simple
>> but moving the iteration over sets to covoar will probably be faster.
>>
>> I think DesiredSymbols is instanced a single time. As a starting point,
>> there
>> would have to be multiple instances of this -- one for each symbol set.
>>
>> Beyond that, there is a correlation between the report generated and
>> the desired symbols set.
>>
>> So I am thinking that we need to define a "context" structure for each
>> set. One simple thought is that there is one "master/unified"
>> DesiredSymbol
>> set under the hood and the symbols in each set are used as a filter
>> when generating reports. So process the executables for every symbol
>> but generate the report subdirectories based on one of the sets of
>> DesiredSymbols.
>>
>> I think that should work.
>>
>> Cillian.. you have been through the flow. Am I thinking right that it is
>>
>> And I think we need to merge before doing this type of work. If we can
>> process
>> a single set correctly, that's a baseline. Adding iteration will be
>> easier to review
>> as another patch.
>>
>>
> we can process a single set correctly.
> Shall we proceed with merging the above patch with the suggested small
> edits
> and then file a ticket for the iteration and then start working on it ?
>

Please.

>
> I am confused with running of coverage for multiple bsp. If a single
> report.html has to show the reports of multiple bsps (as hinted by Cillian)
> Then a lot of rewroking would be needed. If we're headed in that way
> then I think making separate report.html like leon3-report.html would be
> simpler
> to achieve, and then create a master index.html for listing all the report
> htmls.
>

A single run of the tester will test a single BSP. If you open two shell
windows
and run the tester in both, will those collide?

covoar does not need support internally to process multiple BSPs
concurrently.

It is common practice to build and test multiple BSPs in parallel to take
advantage
of machines with many cores.

But if you aren't careful, you can't even have two build/test trees at the
same time
if they collide on file names in shared directories.

Does that make sense?


>> On Mon, Jun 4, 2018 at 1:43 PM, Cillian O'Donnell <cpodonnell8 at gmail.com>
>> wrote:
>>
>>>
>>>
>>> On 4 June 2018 at 19:03, Vijay Kumar Banerjee <vijaykumar9597 at gmail.com>
>>> wrote:
>>>
>>>> On 2 June 2018 at 01:00, Vijay Kumar Banerjee <vijaykumar9597 at gmail.com
>>>> > wrote:
>>>>
>>>>> On 2 June 2018 at 00:48, Joel Sherrill <joel at rtems.org> wrote:
>>>>>
>>>>>>
>>>>>>
>>>>>> On Fri, Jun 1, 2018, 11:21 AM Vijay Kumar Banerjee <
>>>>>> vijaykumar9597 at gmail.com> wrote:
>>>>>>
>>>>>>> On 1 June 2018 at 20:30, Gedare Bloom <gedare at rtems.org> wrote:
>>>>>>>
>>>>>>>> On Fri, Jun 1, 2018 at 10:28 AM, Vijay Kumar Banerjee
>>>>>>>> <vijaykumar9597 at gmail.com> wrote:
>>>>>>>> > On 1 June 2018 at 19:24, Joel Sherrill <joel at rtems.org> wrote:
>>>>>>>> >>
>>>>>>>> >>
>>>>>>>> >>
>>>>>>>> >> On Fri, Jun 1, 2018 at 2:46 AM, Vijay Kumar Banerjee
>>>>>>>> >> <vijaykumar9597 at gmail.com> wrote:
>>>>>>>> >>>
>>>>>>>> >>> Here's the list of Ideas for improvements:
>>>>>>>> >>>
>>>>>>>> >>> 1. Include the section coverage in the bsp config file.
>>>>>>>> >>>     If the section is not found then the script will show
>>>>>>>> >>>     proper error showing coverage is not supported for the
>>>>>>>> >>>     provided bsp config file.
>>>>>>>> >>>
>>>>>>>> >>> 2. Update covoar to add support for separate coverage report
>>>>>>>> >>>     for each symbol set.
>>>>>>>> >>>
>>>>>>>> >>> 3. Add a method somewhere in covoar to get the size of an
>>>>>>>> instruction
>>>>>>>> >>>     and fix the hard coded size 4 in ObjdumpProcessor.cc
>>>>>>>> >>
>>>>>>>> >>
>>>>>>>> >> What about a single XXX_run command? What about that suggestion?
>>>>>>>> >>
>>>>>>>> > The suggestion was to turn test_run and coverage_run into a
>>>>>>>> single command.
>>>>>>>> > I have kept them separate so that there's a possibility to just
>>>>>>>> run the
>>>>>>>> > test.
>>>>>>>> >
>>>>>>>> > If we want to run coverage everytime we run the test. we can do
>>>>>>>> it.
>>>>>>>> > Then I think the --coverage option can be changed to
>>>>>>>> --coverage-sets
>>>>>>>> > to mention the sets.
>>>>>>>> > If that's what we're looking for then I don't think a separate
>>>>>>>> ticket is
>>>>>>>> > needed,
>>>>>>>> > I can try to do it within tomorrow and submit an updated patch.
>>>>>>>> >
>>>>>>>> >>
>>>>>>>> >> Will there be an update to this patch?
>>>>>>>> >>
>>>>>>>> > This patch is working an showing results. I don't have any work
>>>>>>>> > going related to this patch currently.
>>>>>>>> > If there are any suggestions, I'll try to include all the
>>>>>>>> suggested updates
>>>>>>>> > as soon as possible and submit. So that we can get it merged.
>>>>>>>> >
>>>>>>>>
>>>>>>>> I get confused by the similarity between test_run() and
>>>>>>>> coverage_run()
>>>>>>>> names, and now I'm also seeing some confusion because there is a
>>>>>>>> function coverage_run() and a class coverage_run. I suggest you
>>>>>>>> remove
>>>>>>>> this function coverage_run(), and make either
>>>>>>>> coverage_run.__init__()
>>>>>>>> or coverage_run.run() take the executables as a parameter directly.
>>>>>>>>
>>>>>>>> Thank you for the suggestion. :)
>>>>>>> I have removed the function and taken executables as a parameter in
>>>>>>> coverage_run.__init__()
>>>>>>>
>>>>>>> I have a question.
>>>>>>> The ini file that is fed to covoar is written by the script
>>>>>>> according to the
>>>>>>> symbols mentioned by the user. I haven't include the ini file in the
>>>>>>> patch.
>>>>>>> I'm planning to delete the file after the run, unless --no-clean
>>>>>>> option is given,
>>>>>>> which currently deletes the .cov trace files after the run.
>>>>>>>
>>>>>>
>>>>>> That makes sense.
>>>>>>
>>>>>>
>>>>>>> Can I proceed with this ?
>>>>>>>
>>>>>>
>>>>>> Yes.
>>>>>>
>>>>> Added. Thanks. :)
>>>>>
>>>>>> also, shall I include that in the .gitignore ?
>>>>>>>
>>>>>>
>>>>>> Is the name of the file constant? The same across multiple BSPs? If
>>>>>> so, then this will be a problem doing automated testing of multiple BSPs in
>>>>>> parallel.
>>>>>>
>>>>>> The ini file I'm talking about is the symbol sets config file not the
>>>>> bsp
>>>>> config file. yes the name is constant. Should it be unique to the bsp ?
>>>>> something like, leon3-symbols.ini ?
>>>>> How does the automated testing work?
>>>>>
>>>>>> I think the name needs to be unique enough.to support running
>>>>>> testing with coverage on multiple BSPs in parallel. That means you can't
>>>>>> add it to gitignore. And can add another issue and FIXME to the code.
>>>>>>
>>>>>>> If it is needed then I have a fix in mind. I can take the bsp name
>>>>> and add
>>>>> '-symbols.ini' to it. and add *-symbols.ini to .gitignore .
>>>>>
>>>> Shall I add this or put a fixme in the code and post a patch ?
>>>> Are there any other suggestions for the patch ?
>>>>
>>>> I was looking into covoar for generating separate reports for each
>>>> symbolset.
>>>> Seems like  all the coverage reports are generated together and are
>>>> written
>>>> in the _outputDirectory_ .
>>>>
>>>
>>> The output directory should be aligned with the bsp and the report.html
>>> changed to keep record of this instead of searching for the generic
>>> coverage dir. Look for leon3-coverage and so on. As well as the
>>> symbol-sets.ini change you mentioned above. That would probably be enough
>>> to not cause any conflicts with parallel testing (I may be missing a case
>>> there, I let you know if anything else comes to mind). The main thing to
>>> think about is if multiple bsps are being tested at the same time, they
>>> have to know which config file is there's and which output directory and
>>> whatever else they may be looking for, the names have to be unique. These
>>> things are all fed to covoar so the changes can be added to coverage.py.
>>>
>>> I couldn't figure out how to cleanly address this.
>>>> If covoar is intended to generate reports from multiple
>>>> subsystems/symbolsets,
>>>> then I think this would be a needed update. Otherwise we can do it from
>>>> the
>>>> script, by feeding covoar with a single set ini and putting the result
>>>> in a separate
>>>> directory .
>>>>
>>>>> Can we do this ?
>>>>>
>>>>>>
>>>>>>>
>>>>>>>> -Gedare
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> devel mailing list
>>>> devel at rtems.org
>>>> http://lists.rtems.org/mailman/listinfo/devel
>>>>
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20180604/f78127b6/attachment-0002.html>

More information about the devel mailing list