maria-developers team mailing list archive

[Kristian Nielsen <knielsen@xxxxxxxxxxxxxxx>]

"Vladislav Vaintroub" <wlad@xxxxxxxxxxxxxxxx> writes:

> It is my own perception, but I feel the code got severely obfuscated through
> wrapper macros, and thus will be really hard to debug.  Myself, I'd prefer
> it in a C/C++ source file, without macros (yes, it is means repetition, but
> I would not expect the result to be much larger in terms of lines of code
> with this).

Are you sure?

It seems to me a nightmare to maintain the code if the macros were expanded?
Imagine making a change to the wrapper code. Then such change would need to be
made manually in 40 (or whatever) function definitions. It seems this would
very easily lead to subtle errors where one function was incorrectly edited
and so on ...

I mean, sure C macro magic is to be minimised, I agree, but I don't see how?
Or if do you have a concrete suggestion/patch on how to do it better, I'd like
to see it, a version without the macros and without the update/maintenance
problem would indeed be better.

> Also, the macros I talk about define global  functions in a header file. At
> least on Windows  it means linker error (multiple symbol definition) if the
> header is included into 2 different C files.

Here I believe you are talking about nonblock-wrappers.h?

Note that this is only used for test cases in mysqltest.cc and
mysql_client_test.c, it shouldn't be used anywhere else and is not installed
anywhere. So there should be no risk of including multiple times.

This #include could be avoided by instead introducing a .c file and
compiling+linking that in both mysqltest and mysql_client_test. Do you think
that would be better? Or we could even make a libnonblock_wrapper with just
this code, though that seems unnecessarily complex to me, what do you think?

Thanks,

 - Kristian.