[CMake] How to regenerate source file during make?

[Michael Wild]

On 15. Jan, 2010, at 16:38 , David Cole wrote:

> On Fri, Jan 15, 2010 at 10:35 AM, Michael Wild <themiwi at gmail.com> wrote:
> 
>> 
>> On 15. Jan, 2010, at 16:25 , Marcel Loose wrote:
>> 
>>> On Fri, 2010-01-15 at 10:07 -0500, David Cole wrote:
>>>> On Fri, Jan 15, 2010 at 9:44 AM, Marcel Loose <loose at astron.nl> wrote:
>>>>       Hi Jack,
>>>> 
>>>>       The difference with your and my case is that you have an input
>>>>       file
>>>>       (the .xsd file) and output files (the .h and .cpp files). In
>>>>       my case I
>>>>       generate the output file, e.g. myheader.h, from myheader.h.in,
>>>>       using
>>>>       configure_file(). The problem is that I cannot get this to
>>>>       work during
>>>>       'make', only during 'cmake'.
>>>> 
>>>>       Best regards,
>>>>       Marcel Loose.
>>>> 
>>>>       On Fri, 2010-01-15 at 15:04 +0100, Smith Jack (Ext. - UGIS -
>>>>       UniCredit
>>>> 
>>>>       Group) wrote:
>>>>> I do a similar thing with an XSD file from which I generate
>>>>       c++
>>>>       wrapper classes for accessing the corresponding XML
>>>>> 
>>>>> Note the DEPENDS line !
>>>>> 
>>>>> MACRO(XSD FILE NAMESPACE)
>>>>>   ADD_CUSTOM_COMMAND(
>>>>>       OUTPUT  ${CMAKE_CURRENT_SOURCE_DIR}/xml/${FILE}.cpp
>>>>>       OUTPUT  ${CMAKE_CURRENT_SOURCE_DIR}/xml/${FILE}.h
>>>>>       DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/xml/${FILE}.xsd
>>>>>       COMMAND ${XAPI_CODESYNTH}/bin/xsd
>>>>>       ARGS cxx-tree --generate-serialization
>>>>       --hxx-suffix .h
>>>>       --cxx-suffix .cpp --fwd-suffix -fwd.hxx --type-naming java
>>>>       --function-naming java --namespace-map =${NAMESPACE}
>>>>       --generate-polymorphic --output-dir
>>>>       ${CMAKE_CURRENT_SOURCE_DIR}/xml
>>>>       ${CMAKE_CURRENT_SOURCE_DIR}/xml/${FILE}.xsd
>>>>>       COMMENT "generating Codesynthesis Wrappers")
>>>>> ENDMACRO(XSD)
>>>>> 
>>>>> 
>>>>> 
>>>>>> -----Original Message-----
>>>>>> From: cmake-bounces at cmake.org
>>>>>> [mailto:cmake-bounces at cmake.org] On Behalf Of Marcel Loose
>>>>>> Sent: Friday, January 15, 2010 2:55 PM
>>>>>> To: cmake at cmake.org
>>>>>> Subject: [CMake] How to regenerate source file during
>>>>       make?
>>>>>> 
>>>>>> Hi all,
>>>>>> 
>>>>>> I've been struggling with this for hours now and I can't
>>>>       seem
>>>>>> to get a grip on it.
>>>>>> 
>>>>>> I have a header file that must be (re)generated during
>>>>       'make'
>>>>>> when some external data have changed. I can successfully
>>>>>> generate this file during 'cmake' using configure_file(),
>>>>       but
>>>>>> I haven't been able to figure out how I can trigger the
>>>>>> configure_file() during make.
>>>>>> 
>>>>>> Should I define a target for this header file? Or can I
>>>>>> specify some kind of file dependency, such that when the
>>>>>> header file is missing our out-of-date, it will be
>>>>       (re)generated?
>>>>>> 
>>>>>> I guess it must be some clever combination of
>>>>>> add_custom_command() and add_custom_target().
>>>>>> 
>>>>>> Best regards,
>>>>>> Marcel Loose.
>>>> 
>>>> 
>>>> 
>>>> This should work already: if you change my_header.h.in then cmake
>>>> should re-run to re-configure your file at 'make' time.
>>>> 
>>>> 
>>>> If you have additional dependencies that should trigger a
>>>> re-configure, you should probably convert it to an add_custom_command
>>>> such that it happens at build time.
>>> 
>>> It's not the myheader.h.in that's changing; myheader.h should be
>>> regenerated because the value of one of the variables in myheader.h.in
>>> (that get expanded) change.
>>> 
>>> To be more precise, I try to record the svn revision number (and some
>>> more svn info). I carefully read
>>> http://www.mail-archive.com/cmake@cmake.org/msg25922.html, because it
>>> closely matches my problem, but I couldn't figure out what
>>> update_version.cmake should do exactly. In my case, I would need the
>>> values of a number CMake variables (like CMAKE_CURRENT_BINARY_DIR), but
>>> of course, these values are not around anymore when running 'make'.
>>> Should all these variables be passed using -D?
>>> 
>>> Best regards,
>>> Marcel Loose.
>> 
>> You configure_file a CMake script at CMake time which you then use in an
>> add_custom_command to run at make-time. So you put all the "static" data you
>> need (such as CMAKE_BINARY_DIRECTORY, etc.) in the script file and then you
>> have them available at make time:
>> 
>> some_script.cmake:
>> ------------------
>> set(CMAKE_SOURCE_DIR "@CMAKE_SOURCE_DIR@")
>> set(CMAKE_BINARY_DIR "@CMAKE_BINARY_DIR@")
>> set(SVN_EXECUTABLE "@SVN_EXECUTABLE@")
>> # ... do what you need to to ...
>> -------------------
>> 
>> 
>> But I still don't see why you have to extract version-information at build
>> time, because IMHO that shouldn't change between CMake-runs.
>> 
>> 
> I'm assuming he wants to encode "up-to-the-minute" svn status output in the
> build, which changes the second you make a local mod to any source file
> anywhere in your tree...
> 
> Admirable goal, but sure does suck to depend on everything. :-)

Ahhh. I see. That DOES suck. Hope the configured file gets compiled into only one single object file and doesn't trigger an avalanche of recompiling/relinking...

Michael