DOLFIN

Bug #1072694
Comment #9

Comment 9 for bug 1072694

Revision history for this message

Anders Logg (logg) wrote on 2012-10-31: Re: [Bug 1072694] list_timings() cannot be redirected to a file or a variable in python

On Wed, Oct 31, 2012 at 02:53:25PM -0000, Benjamin Kehlet wrote:
> 2012/10/31 corrado maurini <email address hidden>:
> > Johan, I started adding Table output, but this stopped me very soon:
> >
> > As soon I add
> > #include "Table.h"
> > in the log.h I have the following error at compilation:
> >
> > May you give a quick explanation/solution?
>
> I think you introduce a circular dependency of header files here:
>
> log/log.h -> log/Table.h -> common/Variable.h ->
> parameter/Parameters.h -> log/log.h
>
> Maybe class Table doesn't need to inherit class Variable? That would
> solve it, I think, but the core developers should comment on that.

Inheritance from Variable is there so we can do info(table) and info
will know to call table.str.

Circular dependencies can usually be worked around with forward
declarations but I haven't looked at this case in detail.

--
Anders

>
> Regards
> Benjamin
>
> >
> > Corrado
> > -----------------------
> > majorana:dorsal_build_dir maurini$ make all
> > [ 0%] Building CXX object dolfin/CMakeFiles/dolfin.dir/adaptivity/adapt.cpp.o
> > In file included from /opt/HPC/FEniCS/src/dolfin/dolfin/common/Variable.h:27,
> > from /opt/HPC/FEniCS/src/dolfin/dolfin/log/Table.h:28,
> > from /opt/HPC/FEniCS/src/dolfin/dolfin/log/log.h:32,
> > from /opt/HPC/FEniCS/src/dolfin/dolfin/log/dolfin_log.h:6,
> > from /opt/HPC/FEniCS/src/dolfin/dolfin/common/MPI.h:32,
> > from /opt/HPC/FEniCS/src/dolfin/dolfin/fem/UFCCell.h:29,
> > from /opt/HPC/FEniCS/src/dolfin/dolfin/fem/FiniteElement.h:27,
> > from /opt/HPC/FEniCS/src/dolfin/dolfin/adaptivity/adapt.cpp:25:
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h: In member function 'void dolfin::Parameters::add(std::string)':
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:117: error: there are no arguments to 'dolfin_error' that depend on a template parameter, so a declaration of 'dolfin_error' must be available
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:117: error: (if you use '-fpermissive', G++ will accept your code, but allowing the use of an undeclared name is deprecated)
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h: In member function 'void dolfin::Parameters::add(std::string, T, T)':
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:128: error: there are no arguments to 'dolfin_error' that depend on a template parameter, so a declaration of 'dolfin_error' must be available
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h: In member function 'void dolfin::Parameters::add(std::string, std::set<T, std::less<_Key>, std::allocator<_CharT> >)':
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:139: error: there are no arguments to 'dolfin_error' that depend on a template parameter, so a declaration of 'dolfin_error' must be available
> > make[2]: *** [dolfin/CMakeFiles/dolfin.dir/adaptivity/adapt.cpp.o] Error 1
> > make[1]: *** [dolfin/CMakeFiles/dolfin.dir/all] Error 2
> > make: *** [all] Error 2
> >
> > Le 30 oct. 2012 à 21:40, Johan Hake a écrit :
> >
> >> On 10/30/2012 11:53 AM, corrado maurini wrote:
> >>> Johan,
> >>>
> >>> Thanks a lot for taking care of this.
> >>
> >> Well, we see how far it goes :)
> >>
> >>> If redirecting ostream through python is too difficult, for the specific
> >>> case a possible alternative can be to have a function similar to
> >>> list_timings(), but giving a dolfin Table as output.
> >>
> >> Sure that would be pretty trivial to implement, and we have them same
> >> analogy for list_linear_algebra_backends and linear_algebra_backends,
> >> where the later return a list of strings instead of printing them to the
> >> screen.
> >>
> >> So adding a free function called timings would fit fine here.
> >>
> >>> Then the Table is a python object and one can use the associated methods
> >>> to access data or print it out (event in latex format!).
> >>
> >> Good point!
> >>
> >>> I have not done this directly because I do not if this may conflict with
> >>> your general design principles for the logger.
> >>
> >> Should work just fine.
> >>
> >> Johan
> >>
> >>> Corrado
> >>>
> >>> Le 30 oct. 2012 à 11:02, Marco Morandini a écrit :
> >>>
> >>>> On 10/30/2012 08:00 AM, Johan Hake wrote:
> >>>>> Marco, what would the syntax be for adding a file as the log output. Is
> >>>>> it possible to add an already existing file pointer or just the file
> >>>>> name as a string? This would be the ultimate solution as we then can
> >>>>> merge the two log streams from dolfin and logging dependent python
> >>>>> modules such as UFL.
> >>>>>
> >>>>> My boost python skills are really limited so following the inner
> >>>>> workings of this small file is hard for me...
> >>>>>
> >>>>
> >>>> In my old code (where everything is wrapped with boost)
> >>>>
> >>>> after
> >>>>
> >>>> fileout = ostream(cout_buf())
> >>>>
> >>>> fileout will point to std::cout,
> >>>>
> >>>> and after
> >>>>
> >>>> fileout = ostream('new_file_name.txt')
> >>>>
> >>>> fileout will point to a newly opened ofstream.
> >>>>
> >>>> Then, fileout can be passed to any wrapped c++ function that
> >>>> has an std::ostream & argument.
> >>>>
> >>>> I.e., if I have this c++ function
> >>>>
> >>>> void DxOutput(std::ostream & out, const MyFloat t);
> >>>>
> >>>> and I wrap it as
> >>>>
> >>>> BOOST_PYTHON_MEMBER_FUNCTION_OVERLOADS(ofstream_open_overloads,
> >>>> std::ofstream::open, 1, 2);
> >>>> BOOST_PYTHON_MODULE_INIT(_python_util)
> >>>> {
> >>>> ....
> >>>>
> >>>> the wrapping code I've attached
> >>>>
> >>>> ....
> >>>>
> >>>> python::def("DxOutput", DxOutput);
> >>>> }
> >>>>
> >>>> in python I can do
> >>>>
> >>>> t=0.01
> >>>> fileout = ostream('new_file_name.txt')
> >>>> DxOutput(fileout, t)
> >>>>
> >>>> if I want to write with DxOutput into new_file_name.txt,
> >>>>
> >>>> or
> >>>>
> >>>> fileout = ostream(cout_buf())
> >>>> DxOutput(fileout, t)
> >>>>
> >>>> If I want DxOuptut's output on the screen.
> >>>>
> >>>> Note however that I can't write directly from python into fileout
> >>>> (and I fear that this is what you are looking for :( .
> >>>> At least, I've not even tried to do it (perhaps one should start
> >>>> extractig the file descriptor, as explained here:
> >>>>
> >>>> http://www.ginac.de/~kreckel/fileno/
> >>>>
> >>>> )
> >>>>
> >>>> Or
> >>>> http://stackoverflow.com/questions/2378005/c-iostreams-and-python
> >>>>
> >>>>
> >>>> Hope this helps.
> >>>>
> >>>>
> >>>> Title:
> >>>> list_timings() cannot be redirected to a file or a variable in python
> >>>>
> >>>> Status in DOLFIN:
> >>>> Confirmed
> >>>>
> >>>> Bug description:
> >>>> In python it is not possible to redirect the output of list_timings()
> >>>> to a file.
> >>>>
> >>>> set_output_stream(std::ostream& out) in log.ccp is not working in
> >>>> python because of problems in wrapping cpp ostream
> >>>>
> >>>> see also https://answers.launchpad.net/dolfin/+question/211271
> >>>>
> >>>> To manage notifications about this bug go to:
> >>>> https://bugs.launchpad.net/dolfin/+bug/1072694/+subscriptions
> >>>
> >>
> >>
> >> Title:
> >> list_timings() cannot be redirected to a file or a variable in python
> >>
> >> Status in DOLFIN:
> >> Confirmed
> >>
> >> Bug description:
> >> In python it is not possible to redirect the output of list_timings()
> >> to a file.
> >>
> >> set_output_stream(std::ostream& out) in log.ccp is not working in
> >> python because of problems in wrapping cpp ostream
> >>
> >> see also https://answers.launchpad.net/dolfin/+question/211271
> >>
> >> To manage notifications about this bug go to:
> >> https://bugs.launchpad.net/dolfin/+bug/1072694/+subscriptions
> >
> >
> > Title:
> > list_timings() cannot be redirected to a file or a variable in python
> >
> > Status in DOLFIN:
> > Confirmed
> >
> > Bug description:
> > In python it is not possible to redirect the output of list_timings()
> > to a file.
> >
> > set_output_stream(std::ostream& out) in log.ccp is not working in
> > python because of problems in wrapping cpp ostream
> >
> > see also https://answers.launchpad.net/dolfin/+question/211271
> >
> > To manage notifications about this bug go to:
> > https://bugs.launchpad.net/dolfin/+bug/1072694/+subscriptions
>

On Wed, Oct 31, 2012 at 02:53:25PM -0000, Benjamin Kehlet wrote:
> 2012/10/31 corrado maurini <1072694@bugs.launchpad.net>:
> > Johan, I started adding Table output, but this stopped me very soon:
> >
> > As soon I add
> > #include "Table.h"
> > in the log.h I have the following error at compilation:
> >
> > May you give a quick explanation/solution?
>
> I think you introduce a circular dependency of header files here:
>
> log/log.h -> log/Table.h -> common/Variable.h ->
> parameter/Parameters.h -> log/log.h
>
> Maybe class Table doesn't need to inherit class Variable? That would
> solve it, I think, but the core developers should comment on that.

Inheritance from Variable is there so we can do info(table) and info
will know to call table.str.

Circular dependencies can usually be worked around with forward
declarations but I haven't looked at this case in detail.

--
Anders

>
> Regards
> Benjamin
>
> >
> > Corrado
> > -----------------------
> > majorana:dorsal_build_dir maurini$ make all
> > [  0%] Building CXX object dolfin/CMakeFiles/dolfin.dir/adaptivity/adapt.cpp.o
> > In file included from /opt/HPC/FEniCS/src/dolfin/dolfin/common/Variable.h:27,
> >                  from /opt/HPC/FEniCS/src/dolfin/dolfin/log/Table.h:28,
> >                  from /opt/HPC/FEniCS/src/dolfin/dolfin/log/log.h:32,
> >                  from /opt/HPC/FEniCS/src/dolfin/dolfin/log/dolfin_log.h:6,
> >                  from /opt/HPC/FEniCS/src/dolfin/dolfin/common/MPI.h:32,
> >                  from /opt/HPC/FEniCS/src/dolfin/dolfin/fem/UFCCell.h:29,
> >                  from /opt/HPC/FEniCS/src/dolfin/dolfin/fem/FiniteElement.h:27,
> >                  from /opt/HPC/FEniCS/src/dolfin/dolfin/adaptivity/adapt.cpp:25:
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h: In member function 'void dolfin::Parameters::add(std::string)':
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:117: error: there are no arguments to 'dolfin_error' that depend on a template parameter, so a declaration of 'dolfin_error' must be available
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:117: error: (if you use '-fpermissive', G++ will accept your code, but allowing the use of an undeclared name is deprecated)
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h: In member function 'void dolfin::Parameters::add(std::string, T, T)':
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:128: error: there are no arguments to 'dolfin_error' that depend on a template parameter, so a declaration of 'dolfin_error' must be available
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h: In member function 'void dolfin::Parameters::add(std::string, std::set<T, std::less<_Key>, std::allocator<_CharT> >)':
> > /opt/HPC/FEniCS/src/dolfin/dolfin/parameter/Parameters.h:139: error: there are no arguments to 'dolfin_error' that depend on a template parameter, so a declaration of 'dolfin_error' must be available
> > make[2]: *** [dolfin/CMakeFiles/dolfin.dir/adaptivity/adapt.cpp.o] Error 1
> > make[1]: *** [dolfin/CMakeFiles/dolfin.dir/all] Error 2
> > make: *** [all] Error 2
> >
> > Le 30 oct. 2012 à 21:40, Johan Hake a écrit :
> >
> >> On 10/30/2012 11:53 AM, corrado maurini wrote:
> >>> Johan,
> >>>
> >>> Thanks a lot for taking care of this.
> >>
> >> Well, we see how far it goes :)
> >>
> >>> If redirecting ostream through python is too difficult, for the specific
> >>> case a possible alternative can be to have a function similar to
> >>> list_timings(), but giving a dolfin Table as output.
> >>
> >> Sure that would be pretty trivial to implement, and we have them same
> >> analogy for list_linear_algebra_backends and linear_algebra_backends,
> >> where the later return a list of strings instead of printing them to the
> >> screen.
> >>
> >> So adding a free function called timings would fit fine here.
> >>
> >>> Then the Table is a python object and one can use the associated methods
> >>> to access data or print it out (event in latex format!).
> >>
> >> Good point!
> >>
> >>> I have not done this directly because I do not if this may conflict with
> >>> your general design principles for the logger.
> >>
> >> Should work just fine.
> >>
> >> Johan
> >>
> >>> Corrado
> >>>
> >>> Le 30 oct. 2012 à 11:02, Marco Morandini a écrit :
> >>>
> >>>> On 10/30/2012 08:00 AM, Johan Hake wrote:
> >>>>> Marco, what would the syntax be for adding a file as the log output. Is
> >>>>> it possible to add an already existing file pointer or just the file
> >>>>> name as a string? This would be the ultimate solution as we then can
> >>>>> merge the two log streams from dolfin and logging dependent python
> >>>>> modules such as UFL.
> >>>>>
> >>>>> My boost python skills are really limited so following the inner
> >>>>> workings of this small file is hard for me...
> >>>>>
> >>>>
> >>>> In my old code (where everything is wrapped with boost)
> >>>>
> >>>> after
> >>>>
> >>>> fileout = ostream(cout_buf())
> >>>>
> >>>> fileout will point to std::cout,
> >>>>
> >>>> and after
> >>>>
> >>>> fileout = ostream('new_file_name.txt')
> >>>>
> >>>> fileout will point to a newly opened ofstream.
> >>>>
> >>>> Then, fileout can be passed to any wrapped c++ function that
> >>>> has an std::ostream &  argument.
> >>>>
> >>>> I.e., if I have this c++ function
> >>>>
> >>>> void DxOutput(std::ostream & out, const MyFloat t);
> >>>>
> >>>> and I wrap it as
> >>>>
> >>>> BOOST_PYTHON_MEMBER_FUNCTION_OVERLOADS(ofstream_open_overloads,
> >>>> std::ofstream::open, 1, 2);
> >>>> BOOST_PYTHON_MODULE_INIT(_python_util)
> >>>> {
> >>>>     ....
> >>>>
> >>>>       the wrapping code I've attached
> >>>>
> >>>>       ....
> >>>>
> >>>>     python::def("DxOutput", DxOutput);
> >>>> }
> >>>>
> >>>> in python I can do
> >>>>
> >>>> t=0.01
> >>>> fileout = ostream('new_file_name.txt')
> >>>> DxOutput(fileout, t)
> >>>>
> >>>> if I want to write with DxOutput into new_file_name.txt,
> >>>>
> >>>> or
> >>>>
> >>>> fileout = ostream(cout_buf())
> >>>> DxOutput(fileout, t)
> >>>>
> >>>> If I want DxOuptut's output on the screen.
> >>>>
> >>>> Note however that I can't write directly from python into fileout
> >>>> (and I fear that this is what you are looking for :( .
> >>>> At least, I've not even tried to do it (perhaps one should start
> >>>> extractig the file descriptor, as explained here:
> >>>>
> >>>> http://www.ginac.de/~kreckel/fileno/
> >>>>
> >>>> )
> >>>>
> >>>> Or
> >>>> http://stackoverflow.com/questions/2378005/c-iostreams-and-python
> >>>>
> >>>>
> >>>> Hope this helps.
> >>>>
> >>>>
> >>>> Title:
> >>>> list_timings() cannot be redirected to a file or a variable in python
> >>>>
> >>>> Status in DOLFIN:
> >>>> Confirmed
> >>>>
> >>>> Bug description:
> >>>> In python it is not possible to redirect the output of list_timings()
> >>>> to a file.
> >>>>
> >>>> set_output_stream(std::ostream& out) in log.ccp is not working in
> >>>> python because of problems in wrapping cpp ostream
> >>>>
> >>>> see also https://answers.launchpad.net/dolfin/+question/211271
> >>>>
> >>>> To manage notifications about this bug go to:
> >>>> https://bugs.launchpad.net/dolfin/+bug/1072694/+subscriptions
> >>>
> >>
> >>
> >> Title:
> >>  list_timings() cannot be redirected to a file or a variable in python
> >>
> >> Status in DOLFIN:
> >>  Confirmed
> >>
> >> Bug description:
> >>  In python it is not possible to redirect the output of list_timings()
> >>  to a file.
> >>
> >>  set_output_stream(std::ostream& out) in log.ccp is not working in
> >>  python because of problems in wrapping cpp ostream
> >>
> >>  see also https://answers.launchpad.net/dolfin/+question/211271
> >>
> >> To manage notifications about this bug go to:
> >> https://bugs.launchpad.net/dolfin/+bug/1072694/+subscriptions
> >
> >
> > Title:
> >   list_timings() cannot be redirected to a file or a variable in python
> >
> > Status in DOLFIN:
> >   Confirmed
> >
> > Bug description:
> >   In python it is not possible to redirect the output of list_timings()
> >   to a file.
> >
> >   set_output_stream(std::ostream& out) in log.ccp is not working in
> >   python because of problems in wrapping cpp ostream
> >
> >   see also https://answers.launchpad.net/dolfin/+question/211271
> >
> > To manage notifications about this bug go to:
> > https://bugs.launchpad.net/dolfin/+bug/1072694/+subscriptions
>