[CMake] Dependency tracking with constructed source headers...

[Dan Katz]

On 7 Oct 2004, Brad King wrote:

> Dan Katz wrote:
>> So I guess the question is: why didn't the system pick up that
>> foo.o depends on bar.h?  And how can I make it do so?
>
> The dependency scanner for foo.cpp is not currently hooked into the
> set of generated files, so it does not find bar.h until after it is
> generated the first time.  As a bootstrapping mechanism, you can add
> the dependency explicitly:
>
> SET_SOURCE_FILES_PROPERTIES(
> foo.cpp
> PROPERTIES
> OBJECT_DEPENDS ${foo_BINARY_DIR}/bar.h
> )


Thanks for the trick -- it solved my problem quite neatly.  




I noticed, however, that this solution is sort of brittle.  In the
case I posted yesterday, I had a simple dependency tree (generated
files in parens):

                       +--- foo.cpp 
 (foo) <--- (foo.o) <--|  
                       +--- (bar.h)  <---  bar.junk

and everything is fine.  But if I later add another piece to the
project such that I now have a tree like this (modulo the limits of
ASCII art -- there's only one of bar.junk and baz.h, but I can't
figure out a good way to draw the graph conveniently):

                        +--- foo.cpp 
                        |  
         +-- (foo.o) <--+--- (bar.h)  <---  bar.junk
         |              |
         |              +--- baz.h
         |
 (foo) <-+
         |
         |              +--- baz.cpp 
         |              |  
         +-- (baz.o) <--+--- (bar.h)  <---  bar.junk
                        |
                        +--- baz.h

then the build will break if I happen to list baz first in the
list of sources to be built:

  ADD_EXECUTABLE(foo baz foo ${foo_SOURCE_DIR}/bar.junk)

Of course, one either updates the explicit dependency you suggested
above to point to baz.cpp, or simply reorders the source dependencies
in the ADD_EXECUTABLE statement, but it's still not quite as automatic
as one might hope.





Is there a simply way to understand why the dependency tracker doesn't
pick up the dependency on the generated header from the "#include"
statement in the source foo.cpp file?  

How is the system generating the dependencies anyway?  I guess I'll
take a look at the code for the scanner...

Dan