[CMake] CMAKE_MINIMUM_REQUIRED(VERSION 2.6.0 FATAL_ERROR) and cmake_policy(VERSION 2.6.0) not working correctly for 2.6.2

[Brad King]

Alan W. Irwin wrote:
> On 2009-02-01 09:03-0800 Alan W. Irwin wrote:
> 
>> According to the documentation of 2.6.2, policies CMP0008 and CMP0009
>> were introduced after 2.6.0.  Also, according to that documentation,
>> cmake_policy(VERSION 2.6.0) should set those policies to OLD and warn 
>> about
>> that.
>>
>> I use
>>
>> CMAKE_MINIMUM_REQUIRED(VERSION 2.6.0 FATAL_ERROR)
>>
>> according to the documentation that sets
>>
>> cmake_policy(VERSION 2.6.0)
>>
>> by default.  This documented behaviour seems quite logical and useful so
>> that users can use a variety of CMake versions above 2.6.0 and get the
>> same 2.6.0 policies.
>>
>> However, I could see no such warning messages concerning policies 
>> CMP0008 and
>> CMP0009 even with -Wdev (and even if I use cmake_policy(VERSION 2.6.0)
>> explicitly).  I then investigated further with
>>
>> foreach(policy RANGE 0 9)
>>  if(POLICY CMP000${policy})
>>      message(STATUS "Policy ${policy} is ON")
>>  else(POLICY CMP000${policy})
>>      message(STATUS "Policy ${policy} is OFF")
>>  endif(POLICY CMP000${policy})
>> endforeach(policy RANGE 0 9)
>>
>> and obtained
>>
>> -- Policy 0 is ON
>> -- Policy 1 is ON
>> -- Policy 2 is ON
>> -- Policy 3 is ON
>> -- Policy 4 is ON
>> -- Policy 5 is ON
>> -- Policy 6 is ON
>> -- Policy 7 is ON
>> -- Policy 8 is ON
>> -- Policy 9 is ON
>>
>> I am pretty sure the ON result for policies CMP0008 and CMP0009 for
>>
>> CMAKE_MINIMUM_REQUIRED(VERSION 2.6.0 FATAL_ERROR)
>>
>> must be a bug in 2.6.2.
> 
> Oops!  That test simply asked if a policy exists.  So that test is 
> ambiguous
> at best.  Here is an improved test and results:
> 
> foreach(policy RANGE 0 9)
>   cmake_policy(GET CMP000${policy} policy_result)
>   message(STATUS "Policy CMP000${policy} is ${policy_result}")
> endforeach(policy RANGE 0 9)
> 
> -- Policy CMP0000 is NEW
> -- Policy CMP0001 is NEW
> -- Policy CMP0002 is NEW
> -- Policy CMP0003 is NEW
> -- Policy CMP0004 is NEW
> -- Policy CMP0005 is NEW
> -- Policy CMP0006 is NEW
> -- Policy CMP0007 is NEW
> -- Policy CMP0008 is -- Policy CMP0009 is
> 
> So from these results (and if you can trust the documentation of
> cmake_policy(GET ...)) it appears CMP0008 and CMP0009 are not set.  
> Shouldn't
> those be set to OFF with a warning (i.e., the documented behaviour) when

Everything is behaving as intended and as documented.  You're just 
misinterpreting the documentation.  See below.

> CMAKE_MINIMUM_REQUIRED(VERSION 2.6.0 FATAL_ERROR)
> 
> is used?
> 
> If a policy is undefined as for CMP0008 and CMP0009 for 2.6.2 what does 
> that
> mean? Is that the same as OLD (or NEW) or does the policy implementation
> have buggy/unknown behaviour under these circumstances?

The policies are defined but "unset".  I've updated the documentation to 
clarify this:

- "All policies introduced after the specified version will be reset "
- "to use OLD behavior with a warning.  "
+ "All policies introduced after the specified version will be unset.  "

The documentation of cmake_poilcy already says

"When CMake needs to know which behavior
  to use it checks for a setting specified by the project.  If no
  setting is available the OLD behavior is assumed and a warning is
  produced requesting that the policy be set."

Therefore you'll only see a warning if CMake encounters a situation in 
which it needs to know which behavior to use.  In CMP0009 for example, 
you should see a warning only if file(GLOB) encounters a symlink.

-Brad