Boost.test applicability
I have a alrge number of numeric functions (~300) overloaded on various types. What I want to do is having some automated test that check if those functions returns proper values and behave correctly in term of precision. I get a look at Boost.Test but i can't really see if it's applicable to my need and, if it is, how exactly am I supposed to generate that much test and cover all cases (random value testing, singular value test, nan/inf handling etc...). My first idea was to use Boost.preprcoessor file iteration to quickly generate BOOST_TEST_CASE functions but I'm not sure this is a scalable practice. Other point, I saw that there is check in Boost.test which check for an arbitrary precision result. However it uses raw value (like 0.001) where having a check based on ULP or RMS maybe better. Is there anyway to integrate those test using existing test framework ? Thanks in advance
Joel Falcou
I have a alrge number of numeric functions (~300) overloaded on various types. What I want to do is having some automated test that check if those functions returns proper values and behave correctly in term of
How automated? Are you going to write these assertions?
precision. I get a look at Boost.Test but i can't really see if it's applicable to my need and, if it is, how exactly am I supposed to generate that much test and cover all cases (random value testing, singular value test, nan/inf handling etc...).
Boost.Test by itself do not have test suite of checks to validate your factions. At the end of the day you are the one who will have to implement them. Boost.Test does provides some help for you though to simplify the task.
My first idea was to use Boost.preprcoessor file iteration to quickly generate BOOST_TEST_CASE functions but I'm not sure this is a scalable practice.
You can probably use Boost.PP, but I'd recommend looking into Boost.Test's BOOST_AUTO_TEST_CASE_TEMPLATE (http://tinyurl.com/n28zj8) This should allow you to write single test case template and run it with a collection of different types (if that what you need). You can also look into BOOST_PARAM_TEST_CASE (http://tinyurl.com/maq4tj) which will allow you to run the test assertions against different runtime parameters. As for the loating point comparison tools, Boost.Test provides you several of those: BOOST_CHECK_CLOSE, BOOST_CHECK_CLOSE_FRACTION, BOOST_CHECK_SMALL (see reference http://www.boost.org/doc/libs/1_39_0/libs/test/doc/html/utf/testing-tools/re... for more info ) They follow slightly different logic in selecting tolerance.
Other point, I saw that there is check in Boost.test which check for an arbitrary precision result. However it uses raw value (like 0.001) where having a check based on ULP or RMS maybe better. Is there anyway to integrate those test using existing test framework ?
There was a discussion about adding a tool that checks based on number of ULP. For now you can mimic it yourself with proper tolerance specification. Not sure what you mean by RMS, Regards, Gennadiy
Gennadiy Rozental wrote:
How automated? Are you going to write these assertions? I was thinkign to BOOST_PP them. Boost.Test by itself do not have test suite of checks to validate your factions. At the end of the day you are the one who will have to implement them. Boost.Test does provides some help for you though to simplify the task.
Yeah I figured it out later myself. Sorry for the noise
You can probably use Boost.PP, but I'd recommend looking into Boost.Test's BOOST_AUTO_TEST_CASE_TEMPLATE (http://tinyurl.com/n28zj8)
OK
There was a discussion about adding a tool that checks based on number of ULP. This is very sensible. I may provide th code if needed as I was doing it by hand before
Not sure what you mean by RMS,
RMS = root-meaned squared http://en.wikipedia.org/wiki/Root_mean_square -- ___________________________________________ Joel Falcou - Assistant Professor PARALL Team - LRI - Universite Paris Sud XI Tel : (+33)1 69 15 66 35
joel
RMS = root-meaned squared http://en.wikipedia.org/wiki/Root_mean_square
And how can it be used to compare floating point numbers?
Joel Falcou wrote:
Other point, I saw that there is check in Boost.test which check for an arbitrary precision result. However it uses raw value (like 0.001) where having a check based on ULP or RMS maybe better. Is there anyway to integrate those test using existing test framework ?
So, how did you mean to use RMS for FP value comparison? Gennadiy
participants (3)
-
Gennadiy Rozental -
joel -
Joel Falcou