TestPlugin: Avoid generating warnings in client codes from old style …

[bukulin]

…casts

When -Wold-style-cast is enabled

[basvodde]

We avoid new casts due to old compilers. However, the failure from the old Watcom compiler is interesting. Did you look at that?

[bukulin]

Shame on me! I appologize about the early pull request. I will look into it during the day.

However, I think I was not clear enough in the short description. Sorry about that. So my problem is, if I turn on the -Wold-style-cast warning flag in my project, then I get warnings about old style casts generated by the UT_PTR_SET macro.

If I understand you correctly, the acceptable solution for C++ is a function template instead of a new reinterpret_cast. Is that the correct way?

[basvodde]

Uhm, no worry about early pull requests. It is good as it allows for discussion.

Templates are not good either. Because CppUTest is used in some embedded products, we try to keep the level of C++ used to be 'old' so that we support these compilers. So, in CppUTest itself, I don't think we use the 'new' C++ style casts. I would be tempted to merge it (I understand your problem) but it seems the travis build failed because of it. I don't understand very well why, but you might wanna check that first.

[coveralls]

Coverage remained the same at 99.876% when pulling b34b923 on bukulin:master into 7562465 on cpputest:master.

[arstrube]

It failed linking. There doesn't seem to be any problem with any code.

I suspect one binary perhaps grew too large. The tests are distributed over several binaries, and it is possible for one of them to break the +/- 640k limit.

[arstrube]

Sorry. Looking at it a second time, I see a problem with the code :-(

[arstrube]

.../CheatSheetTest.cpp(15): Error! E784: col(34) reinterpret_cast expression must be pointer or integral object
.../CheatSheetTest.cpp(15): Note! N630: col(34) source conversion type is 'void (* * )( void )'
.../CheatSheetTest.cpp(15): Note! N631: col(34) target conversion type is 'void * *'
Error: Compiler returned a bad status compiling '/home/travis/build/cpputest/cpputest/tests/CppUTest/CheatSheetTest.cpp'

[bukulin]

I had an assumption about a vexing parse, but not that is the situation. I think, maybe the Watcom compiler handles pointers to functions and pointers to variables slightly differently.
Unfortunately I do not have any experience with the Watcom compiler, but I try to figure it out.

[arstrube]

I wish I could help you there. But even though it was me who added the Watcom stuff (to be able to test some form of 16 bit code), much of my work with Watcom was by trial and error.

[basvodde]

I'm surprised the compiler has support for reinterpret_cast :)

[arstrube]

An old architecture, but the compiler has been kept up-to-date ;-)

[basvodde]

Oh, has it. We'll then need to add a really old compiler to the build too.

[arstrube]

Perhaps we need to redefine "old" ?

[arstrube]

Sometimes companies are required to be able to build firmware from, say, 20 years ago. They have kept the old code. They even have a copy of the old build tools archived somewhere. But sadly, they have no more machine to run it on... And this is really happening! :-)

[Andne]

I know what you mean, have a VM at work to run one of our old compilers on since the installer can't even start on most of our computers anymore. We're still using that compiler on active projects too unfortunately, don't know when it will truly become retired.

Even if someone resurrects an old project with an older compiler, wouldn't they be trying to use the version of CppUTest that was around back then? In my opinion, the unit test framework is effectively one of the build tools. I understand maintaining compatibility, but seems like that eventually becomes too much of a burden and the supported environment should be updated to be one that can support newer functionality.

Specific to this issue, I do like the new-style casts myself, the different ones makes it a bit easier to avoid accidental effects by ensuring that you can only perform certain types of casting within each one.

[bukulin]

Actually, I don't see any other solution for the problem then a big, fat chunk of #ifdef-s, but I do not like that at all. Do you have any other idea?

[basvodde]

@bukulin An #if might work, especially for public interfaces.

However, it would be good to first figure out the Watcom thing. If it is due to no support for reinterpretted_cast then I guess it would be a good way to check. But I'm not quite sure why the Watcom compiler is failing.