[cmake-developers] Topic hdf5-module-bug-fix

[Brad King]

On 05/24/2011 03:04 PM, Will Dicharry wrote:
> Brad King wrote:
>> Therefore if someone finds HDF5 twice the second time may take the wrong
>> logic path.  The proper value to test after the find_package call is
>> HDF5_FOUND to know whether or not the config file was found and loaded.
> 
> Won't this have the same problem? HDF5_FOUND gets set by 
> FindPackageHandleStandardArgs, so the second time find_package(HDF5) 
> gets called on a system with the autoconf HDF5 build we'll take the 
> wrong control path and call get_target_properties on targets that don't 
> exist.

Yes, you're right.

> That's why I used the INCLUDE_DIR in the first place (forgetting 
> about the backwards compatibility thing at the end).  Is there something 
> else I can branch on?

The find_package command will set HDF5_DIR to the directory containing
the config file.  However, there is a different organization we can use.

The general form should be something like

 if(NOT HDF5_FOUND)
   find_package(HDF5 QUIET NO_MODULE)
   if(HDF5_FOUND)
     # ... map hdf5-config values to our outputs
   endif()
 endif()

 if(NOT HDF5_FOUND)
   # old-style search
 endif()

 if(HDF5_FOUND)
   # Generic logic (if any) when HDF5 is found by either method.
   # Logic should be safe to run twice or have its own guard.
 endif()

By including the "NOT HDF5_FOUND" guard around each block then it does
not matter how many times the module is included.

-Brad