On 3/21/2016 4:55 AM, Florian Lindner wrote:

Hey,

snipped ...

I also want to write some text between the sequence items, I have made it working with that code:

#define PRINT(r, data, elem) \ cout << "Argument " << r-1 << ": " << elem << endl;

#define vmacro(expr, ...) \ cout << "Expression: " << #expr << endl; \ BOOST_PP_SEQ_FOR_EACH(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \

So, invoking vmacro(1-2, 5, 6, "Hallo"); prints

#define PRINT(r, data, elem) \ cout << "Argument " << r-1 << ": " << elem << endl;

#define vmacro(expr, ...) \ cout << "Expression: " << #expr << endl; \ BOOST_PP_SEQ_FOR_EACH(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \

Expression: 1-2 Argument 1: 5 Argument 2: 6 Argument 3: Hallo

Any comments on that? How to do it better? Maybe without a helper macro PRINT? I'm a bit surprised I have to use r-1 to get 1 for the first argument.

The PRINT macro is fine. Almost all Boost PP looping constructs use a helper macro like that. In real life code I would name all macros in uppercase and give them a very distinct name to reduce clashes with other potential C++ identifiers and macros. Look at BOOST_PP_SEQ_FOR_EACH_I to get the actual argument index instead of using 'r'. That 'r' in your code above is actually 2,3, and 4 is just luck. The 'next available BOOST_PP_FOR repetition' as described in the doc just happened to be 2,3, and 4 but could theoretically have been any PP number each time and is not related to the index of each argument.

On 3/22/2016 3:35 PM, Florian Lindner wrote:

On Mon, 21 Mar 2016 10:39:52 -0400 Edward Diener wrote:

...

On 3/21/2016 4:55 AM, Florian Lindner wrote:

...

Hey,

snipped ...

...

[...]

...

Any comments on that? How to do it better? Maybe without a helper macro PRINT? I'm a bit surprised I have to use r-1 to get 1 for the first argument.

The PRINT macro is fine. Almost all Boost PP looping constructs use a helper macro like that.

In real life code I would name all macros in uppercase and give them a very distinct name to reduce clashes with other potential C++ identifiers and macros.

Sure!

...

Look at BOOST_PP_SEQ_FOR_EACH_I to get the actual argument index instead of using 'r'.

That 'r' in your code above is actually 2,3, and 4 is just luck. The 'next available BOOST_PP_FOR repetition' as described in the doc just happened to be 2,3, and 4 but could theoretically have been any PP number each time and is not related to the index of each argument.

Ok, try to use FOR_EACH_I, but ran into compilation errors:

#define PRINT(r, data, i, elem) \ cout << "Argument " << i << ": " << elem << endl;

#define vmacro(expr, ...) \ cout << "Expression: " << #expr << endl; \ BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \

What is that final '\' about ? Please show the header files being included.

but:

% g++ -std=c++11 -g3 -rdynamic -ldl test.cpp && ./a.out :( test.cpp: In function 'void f(int)': test.cpp:127:27: error: 'PRINT' was not declared in this scope BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^ test.cpp:127:34: error: expected primary-expression before ',' token BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^ In file included from /usr/include/boost/preprocessor/variadic/to_seq.hpp:17:0, from test.cpp:11: test.cpp:127:36: error: expression cannot be used as a function BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^ test.cpp:127:73: error: 'BOOST_PP_SEQ_FOR_EACH_I' was not declared in this scope BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^

boost 1.60, gcc 5.3.0.

The same code, using FOR_EACH works (only the argumentens of PRINT were altered.

Thanks!

Florian

Hello, another question: I was expecting that the line BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); in a macro #define vm(expr, ...) expands to nothing, when the the macro is called with just one argument vm(false), but it fails, only when called with just one argument: % g++ -std=c++11 -g3 test.cpp test.cpp: In function 'void f(int)': test.cpp:109:51: error: expected primary-expression before '<<' token std::cerr << " Argument " << i << ": " << elem << std::endl; ^ /usr/include/boost/preprocessor/seq/for_each_i.hpp:85:66: note: in expansion of macro 'PRINT_ARGUMENT' # define BOOST_PP_SEQ_FOR_EACH_I_M_I(r, macro, data, seq, i, sz) macro(r, data, i, BOOST_PP_SEQ_HEAD(seq)) ^ test.cpp:140:1: note: in expansion of macro 'assertion' assertion(!true); ^ Is there any way to make it work like that? Thanks a lot! Florian The code looks like that: #define PRINT_ARGUMENT(r, data, i, elem) \ std::cerr << " Argument " << i << ": " << elem << std::endl; #define assertion(expr, ...) if (!(expr)) { \ std::cerr << "Assertion in " << __FILE__ << ":" << __LINE__ \ << ", failed expression: " << #expr << std::endl; \ std::cout << BOOST_PP_VARIADIC_SIZE(__VA_ARGS__) << std::endl; \ BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ std::cerr.flush(); \ std::cout.flush(); \ assert(false); \ } \ On Tue, 22 Mar 2016 20:53:06 -0400 Edward Diener wrote:

On 3/22/2016 3:35 PM, Florian Lindner wrote:

...

On Mon, 21 Mar 2016 10:39:52 -0400 Edward Diener wrote:

...

On 3/21/2016 4:55 AM, Florian Lindner wrote:

...

Hey,

snipped ...

...

[...]

...

Any comments on that? How to do it better? Maybe without a helper macro PRINT? I'm a bit surprised I have to use r-1 to get 1 for the first argument.

The PRINT macro is fine. Almost all Boost PP looping constructs use a helper macro like that.

In real life code I would name all macros in uppercase and give them a very distinct name to reduce clashes with other potential C++ identifiers and macros.

Sure!

...

Look at BOOST_PP_SEQ_FOR_EACH_I to get the actual argument index instead of using 'r'.

That 'r' in your code above is actually 2,3, and 4 is just luck. The 'next available BOOST_PP_FOR repetition' as described in the doc just happened to be 2,3, and 4 but could theoretically have been any PP number each time and is not related to the index of each argument.

Ok, try to use FOR_EACH_I, but ran into compilation errors:

#define PRINT(r, data, i, elem) \ cout << "Argument " << i << ": " << elem << endl;

#define vmacro(expr, ...) \ cout << "Expression: " << #expr << endl; \ BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \

What is that final '\' about ?

Please show the header files being included.

...

but:

% g++ -std=c++11 -g3 -rdynamic -ldl test.cpp && ./a.out :( test.cpp: In function 'void f(int)': test.cpp:127:27: error: 'PRINT' was not declared in this scope BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^ test.cpp:127:34: error: expected primary-expression before ',' token BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^ In file included from /usr/include/boost/preprocessor/variadic/to_seq.hpp:17:0, from test.cpp:11: test.cpp:127:36: error: expression cannot be used as a function BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^ test.cpp:127:73: error: 'BOOST_PP_SEQ_FOR_EACH_I' was not declared in this scope BOOST_PP_SEQ_FOR_EACH_I(PRINT, , BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__)); \ ^ test.cpp:159:3: note: in expansion of macro 'vmacro' vmacro(1-2, 5, 6); ^

boost 1.60, gcc 5.3.0.

The same code, using FOR_EACH works (only the argumentens of PRINT were altered.

Thanks!

Florian

On Wed, 23 Mar 2016 11:56:25 -0400 Edward Diener wrote:

On 3/23/2016 11:45 AM, Florian Lindner wrote:

...

Hello,

another question: I was expecting that the line

BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__));

in a macro #define vm(expr, ...) expands to nothing, when the the macro is called with just one argument vm(false), but it fails, only when called with just one argument:

% g++ -std=c++11 -g3 test.cpp

test.cpp: In function 'void f(int)': test.cpp:109:51: error: expected primary-expression before '<<' token std::cerr << " Argument " << i << ": " << elem << std::endl; ^ /usr/include/boost/preprocessor/seq/for_each_i.hpp:85:66: note: in expansion of macro 'PRINT_ARGUMENT' # define BOOST_PP_SEQ_FOR_EACH_I_M_I(r, macro, data, seq, i, sz) macro(r, data, i, BOOST_PP_SEQ_HEAD(seq)) ^ test.cpp:140:1: note: in expansion of macro 'assertion' assertion(!true); ^

Is there any way to make it work like that?

When you specify '...' to indicate variadic arguments you must always supply at least one argument. That's part of the C++ standard.

The way to make 'vm' work is to specify:

#define vm(...)

and then extract your first argument and pass the remaining arguments, if they exist, to your BOOST_PP_SEQ_FOR_EACH_I expansion. I will let you figure out how to do that, but if you find you can't figure it out post back and I will show you the code.

Hey, ok, I used SEQ_TAIL for that and it works nicely: #define PRINT_ARGUMENT(r, data, i, elem) \ std::cerr << " Argument " << i << ": " << elem << std::endl; #define assertion(...) if (not BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) { \ std::cerr << "Assertion in " << __FILE__ << ":" << __LINE__ \ << ", failed expression: " << BOOST_PP_STRINGIZE(BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) << std::endl; \ BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_SEQ_TAIL(BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__))); \ std::cerr.flush(); \ std::cout.flush(); \ assert(false); \ } as you can see, it's going to be an asssert macro. STRINGIZE gives me the literal expression, like the # operator. Any comments? Best Thanks, Florian

On Thu, 24 Mar 2016 11:00:57 -0400 Edward Diener wrote:

On 3/24/2016 4:42 AM, Florian Lindner wrote:

...

On Wed, 23 Mar 2016 11:56:25 -0400 Edward Diener wrote:

...

On 3/23/2016 11:45 AM, Florian Lindner wrote:

...

Hello,

another question: I was expecting that the line

BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__));

in a macro #define vm(expr, ...) expands to nothing, when the the macro is called with just one argument vm(false), but it fails, only when called with just one argument:

% g++ -std=c++11 -g3 test.cpp

test.cpp: In function 'void f(int)': test.cpp:109:51: error: expected primary-expression before '<<' token std::cerr << " Argument " << i << ": " << elem << std::endl; ^ /usr/include/boost/preprocessor/seq/for_each_i.hpp:85:66: note: in expansion of macro 'PRINT_ARGUMENT' # define BOOST_PP_SEQ_FOR_EACH_I_M_I(r, macro, data, seq, i, sz) macro(r, data, i, BOOST_PP_SEQ_HEAD(seq)) ^ test.cpp:140:1: note: in expansion of macro 'assertion' assertion(!true); ^

Is there any way to make it work like that?

When you specify '...' to indicate variadic arguments you must always supply at least one argument. That's part of the C++ standard.

The way to make 'vm' work is to specify:

#define vm(...)

and then extract your first argument and pass the remaining arguments, if they exist, to your BOOST_PP_SEQ_FOR_EACH_I expansion. I will let you figure out how to do that, but if you find you can't figure it out post back and I will show you the code.

Hey,

ok, I used SEQ_TAIL for that and it works nicely:

That's the solution.

I could add a BOOST_PP_VARIADIC_TAIL but keeping the Boost PP variadic macros to a minimum was the idea when they were created.

...

#define PRINT_ARGUMENT(r, data, i, elem) \ std::cerr << " Argument " << i << ": " << elem << std::endl;

#define assertion(...) if (not BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) { \

Wouldn't

if !(BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) { \

be better ?

The if condition needs to be enclosed by parens, don't it? This piece of code does not compile? Or do you propose to replace the "not" by "!"? I think they are the same, I just like the "not" better, since it's more literal.

...

std::cerr << "Assertion in " << __FILE__ << ":" << __LINE__ \ << ", failed expression: " << BOOST_PP_STRINGIZE(BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) << std::endl; \ BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_SEQ_TAIL(BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__))); \ std::cerr.flush(); \ std::cout.flush(); \ assert(false); \ }

as you can see, it's going to be an asssert macro. STRINGIZE gives me the literal expression, like the # operator.

Any comments?

I'm glad Boost PP is useful to you <g>. If you are further interested in preprocessor programming in general you might find my Boost VMD library interesting ( shameless plug ) <g>.

Thanks, I'll keep it in mind. But I try to keep preprocessor programming to a minimum. Best, Florian

Hey, I have integrated the code in our project now and it fails to compile with some older clang version. It compiles fine with clang++ 3.7.1 and g++ 5.3.0 on my arch machine as well as g++ 4.8.4 on our ubuntu server, but fails with clang++ 3.4 on this machine as well as the travis The complete compiler messages are at http://pastebin.com/HgjgCfsH it is mostly something like that: error: unknown type name 'BOOST_PP_IIF_0' assertion( reinterpret_cast(*this).size()==toAssign.size() ); The assertions code is at https://github.com/precice/precice/blob/develop/src/utils/assertion.hpp as well as all other code files. If you want to give it a try, you can clone the project and use clang++ -o build/debug-nopetsc-nompi/tarch/configuration/ConfigurationRegistry.o -c -Wall -std=c++11 -fPIC -Wsign-compare -g3 -O0 -DDebug -DAsserts -DPRECICE_NO_PETSC -DEIGEN_INITIALIZE_MATRICES_BY_NAN -DPRECICE_NO_MPI -DNPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION -Isrc -I/usr/include -I/usr/include/python2.7 -I/usr/include/python2.7/numpy src/tarch/configuration/ConfigurationRegistry.cpp (it's also included in the pastebin above) Do you have any idea? Best, Florian On Tue, 29 Mar 2016 09:32:54 -0400 Edward Diener wrote:

On 3/29/2016 6:49 AM, Florian Lindner wrote:

...

On Thu, 24 Mar 2016 11:00:57 -0400 Edward Diener wrote:

...

On 3/24/2016 4:42 AM, Florian Lindner wrote:

...

On Wed, 23 Mar 2016 11:56:25 -0400 Edward Diener wrote:

...

On 3/23/2016 11:45 AM, Florian Lindner wrote:

...

Hello,

another question: I was expecting that the line

BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__));

in a macro #define vm(expr, ...) expands to nothing, when the the macro is called with just one argument vm(false), but it fails, only when called with just one argument:

% g++ -std=c++11 -g3 test.cpp

test.cpp: In function 'void f(int)': test.cpp:109:51: error: expected primary-expression before '<<' token std::cerr << " Argument " << i << ": " << elem << std::endl; ^ /usr/include/boost/preprocessor/seq/for_each_i.hpp:85:66: note: in expansion of macro 'PRINT_ARGUMENT' # define BOOST_PP_SEQ_FOR_EACH_I_M_I(r, macro, data, seq, i, sz) macro(r, data, i, BOOST_PP_SEQ_HEAD(seq)) ^ test.cpp:140:1: note: in expansion of macro 'assertion' assertion(!true); ^

Is there any way to make it work like that?

When you specify '...' to indicate variadic arguments you must always supply at least one argument. That's part of the C++ standard.

The way to make 'vm' work is to specify:

#define vm(...)

and then extract your first argument and pass the remaining arguments, if they exist, to your BOOST_PP_SEQ_FOR_EACH_I expansion. I will let you figure out how to do that, but if you find you can't figure it out post back and I will show you the code.

Hey,

ok, I used SEQ_TAIL for that and it works nicely:

That's the solution.

I could add a BOOST_PP_VARIADIC_TAIL but keeping the Boost PP variadic macros to a minimum was the idea when they were created.

...

#define PRINT_ARGUMENT(r, data, i, elem) \ std::cerr << " Argument " << i << ": " << elem << std::endl;

#define assertion(...) if (not BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) { \

Wouldn't

if !(BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) { \

be better ?

The if condition needs to be enclosed by parens, don't it? This piece of code does not compile?

Yes, of course. My error ( wipes egg off face ).

...

Or do you propose to replace the "not" by "!"? I think they are the same, I just like the "not" better, since it's more literal.

OK. I rarely use the alphabetic alternatives to the logical operators.

...

...

...

std::cerr << "Assertion in " << __FILE__ << ":" << __LINE__ \ << ", failed expression: " << BOOST_PP_STRINGIZE(BOOST_PP_VARIADIC_ELEM(0, __VA_ARGS__)) << std::endl; \ BOOST_PP_SEQ_FOR_EACH_I(PRINT_ARGUMENT,, BOOST_PP_SEQ_TAIL(BOOST_PP_VARIADIC_TO_SEQ(__VA_ARGS__))); \ std::cerr.flush(); \ std::cout.flush(); \ assert(false); \ }

as you can see, it's going to be an asssert macro. STRINGIZE gives me the literal expression, like the # operator.

Any comments?

I'm glad Boost PP is useful to you <g>. If you are further interested in preprocessor programming in general you might find my Boost VMD library interesting ( shameless plug ) <g>.

Thanks, I'll keep it in mind. But I try to keep preprocessor programming to a minimum.

Sure.

There are some really neat things you can do using the preprocessor using Boost PP, and Boost VMD extends that realm when variadic parameters are available, as they are for all modern compilers.

Both libraries are aids for more complicated C++ macro writing than normal.

On Wed, 30 Mar 2016 14:42:41 -0400 Edward Diener wrote:

On 3/30/2016 4:35 AM, Florian Lindner wrote:

...

Hey,

I have integrated the code in our project now and it fails to compile with some older clang version.

It compiles fine with clang++ 3.7.1 and g++ 5.3.0 on my arch machine as well as g++ 4.8.4 on our ubuntu server, but fails with clang++ 3.4 on this machine as well as the travis

The complete compiler messages are at http://pastebin.com/HgjgCfsH it is mostly something like that:

error: unknown type name 'BOOST_PP_IIF_0' assertion( reinterpret_cast(*this).size()==toAssign.size() );

The assertions code is at https://github.com/precice/precice/blob/develop/src/utils/assertion.hpp

as well as all other code files.

If you want to give it a try, you can clone the project and use

clang++ -o build/debug-nopetsc-nompi/tarch/configuration/ConfigurationRegistry.o -c -Wall -std=c++11 -fPIC -Wsign-compare -g3 -O0 -DDebug -DAsserts -DPRECICE_NO_PETSC -DEIGEN_INITIALIZE_MATRICES_BY_NAN -DPRECICE_NO_MPI -DNPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION -Isrc -I/usr/include -I/usr/include/python2.7 -I/usr/include/python2.7/numpy src/tarch/configuration/ConfigurationRegistry.cpp

(it's also included in the pastebin above)

Do you have any idea?

Would you please post a complete example along with your command line as a Boost bug report for the preprocessor library ? That way I am more likely to take a look at it. My guess, without knowing the details of your example, is just a clang-3.4 bug which has been fixed in a later release but if your problem is easily reproducable I can add a test to the preprocessor tests, based on your example, which shows the clang-3.4 bug. Currently my preprocessor tests using clang-3.4 do not show any problems so incorporating such a test showing a problem will be worthwhile for me.

Hey, I'm very happy to help that way. I was able to reproduce that bug in a small piece of code and was doing some tests using different versions. The bug is triggered using all clang versions (3.4 and 3.7.1 at least) if using boost 1.55, and obviously was fixed in 1.56. Do you still want me to open a bug report? Best, Florian