Re: [Cppunit-devel] CppUnitW 1.2 merge

[Baptiste L. <bl...@cl...>]

> I've merged most of CppUnitW 1.2, with some exceptions:
    Thanks, this is excellent !!!

>
> * TestFactoryRegistry used auto_ptr. I had to change that to ordinary
> pointers: auto_ptr's cannot be used inside STL containers (see for
> example:  http://gcc.gnu.org/onlinedocs/libstdc++/20_util/howto.html ).
> With MSVC++ it may work (by accident), but with g++ is doesn't even
> compile. For now, I've only changed pointer types and not done anything
> about the memory management: very likely it will leak now. Please have a
> look at it.
    Thanks for pointing that out. I aware of this limitation (and even with
VC++ your going into trouble if you're using sort(). It's also one of the
reason auto_ptr had such an hard time getting into stl). I did this as a
quick hack to test the thing and completly forgot to clean it up.
    I'll send a patch tomorrow I hope to fix the leaks. Though there is
still some problem when doing:

TestFactoryRegistry::getRegistry().registerFactory( "Functionnal tests",
TestFactoryRegistry::getRegistry( "Functionnal" ) );

Since the registry owns the factories, it will be destroyed twice (owned by
getRegistry( aName ) and registerFactory() ). I'm still turning this around
to see how I can fix that.

>
> * I've put the typeinfo stuff in TestSuite and TestFactoryRegistry
> inside "#ifdef USE_TYPEINFO ".  RTTI is still rather new and may not be
    I would rather use CU_USE_TYPEINFO or something like that
(CU_NO_TYPEINFO would be easier for us since we wouldn't need to declare it
in each project). Prefixes are a great help when you have tens of
precompiler define and you need to figure out which library they are related
to (and with VC++ you a 40 chars edit field to see them all :-( ).

> supported properly on all platforms. Since it's only used to create
> TestSuites with names direved from a class name, I did not not feel
> dependence of the core library on RTTI was justified. Also deriving the
    I agree with that. since it has been available with VC++ for years (it
was there in VC++ 5.0), I though it was a rather comon feature by now. Well
I guessed wrong... ;-(

> class name through type_info is an unreliable method. Once again the C++
> comittee managed to kill a useful feature: not only may the string
> returned by type_info.name() differ from compiler to compiler (e.g. g++
> returns some really ugly mangled name), but differ with different runs
> of the program as well. Of course this renders the whole method useful.
    Yukk! Do raw_name() and name() return the same thing (raw_name is not
supposed to be human readable, but name() is supposed to be....) ?

> If someone can explain why the comittee has done this, please do!
    Damn. I though this one bit was at least well specified. I guess the use
will mainly be for VC++ user. I'll add another constructor to
TestSuiteBuilder so the suite name can be specified by hand.

>
> * Changed TestSuite.getTests() to return a const ref.
    Well, I find the original one easier to use, but I won't arg about that.

>
> * Moved TestFactoryRegistry.cpp from src/cppunit/extensions to
> src/cppunit. It is part of the main library anyway and is much easier to
> configure automake for it.
    I'll need to fix the VC++ project. Will be in the patch to.
>
> Please give it a try and let me know where I messed up.
    So did I...
    TestAssert.cpp is missing.

    There is also a few minor things (include path in TestRegistry.cpp, typo
in TestRunnerModel.cpp). I'll put those in the patch.

    Other than that it works just fine. Hopefully, I make the patch tomorrow
and we'll soon have a working version !

    Thanks a lot Baastian.

    Baptiste.

---
Baptiste Lepilleur <gai...@fr...>   http://gaiacrtn.free.fr/index.html
Author of The Text Reformatter, a tool for fanfiction readers and writers.
Language: English, French (Well, I'm French).