[CMake] #cmakedefine and the #if vs #ifdef argument

[Hostile Fork]

On Jun 11, 2009, at 14:47 , Pau Garcia i Quiles wrote:
> #define VAR_THAT_IS_ON @VAR_THAT_IS_ON@
> #define VAR_THAT_IS_OFF @VAR_THAT_IS_OFF@
>
> If you call CMake with 'cmake -DVAR_THAT_IS_ON=1 -DVAR_THAT_IS_OFF=0',
> it will produce:
>
> #define VAR_THAT_IS_ON 1
> #define VAR_THAT_IS_OFF 0

Hello Pau, thanks for the quick response...

This approach would work if I were using STRING.  However, the  
variables in question were created using "option", and are  
BOOL...which seems preferable.  Yet this gives rise to some very  
unusual behavior in how they substitute using @VAR at .

Anything equivalent to 'off' (e.g. "0", "off", "falSE", "NO") produces  
an all-caps "OFF".  So that means I get:

	#define VAR_THAT_IS_OFF OFF

If using the command line ("-DVAR_THAT_IS_ON=[stuff]"), then any stuff  
equivalent to 'on' (e.g. "1", "Yes", "oN", "True") is forced to ON.   
So that seemed pretty symmetric:

	#define VAR_THAT_IS_ON ON

I was tempted to just '#define OFF 0' and '#define ON 1' in my file  
and move on.  But when using the interactive mode ("cmake -i") it did  
something unusual... it preserved the specific string you had used to  
specify 'on'.  So you get a wide variety of possibilities, such as:

	#define VAR_THAT_IS_ON 1
	#define VAR_THAT_IS_ON on
	#define VAR_THAT_IS_ON On
	#define VAR_THAT_IS_ON oN
	#define VAR_THAT_IS_ON ON
	#define VAR_THAT_IS_ON yes
	#define VAR_THAT_IS_ON Yes
	#define VAR_THAT_IS_ON yEs
	...
	#define VAR_THAT_IS_ON true
	...
	#define VAR_THAT_IS_ON TRUe
	#define VAR_THAT_IS_ON TRUE

(Note: strings that aren't allowed signifiers of ON/OFF are treated as  
off.  So entering "banana" in the interactive mode will get you an all- 
caps OFF in the substitution.)

The command line does not do the case standardization for variables  
declared as "BOOL", only those made using option.  So if you say:

	set(VAR_BOOL OFF CACHE BOOL "A boolean variable")

Then supplying '-DVAR_BOOL=oN' on the command line will substitute to  
'oN' and not 'ON'.  Changing this to an option gets the normalization  
back:

	option(VAR_BOOL "A boolean variable")


==> SOURCE INVESTIGATION <==

Since I'm curious about CMake and understanding what it does, I looked  
into the source.  I gather the short answer is that any current  
appearance of standardization for boolean substitution is unintentional.

I'm not sure how valuable establishing such a standard would be...  
since it could apply only to 'option' variables (and those 'set'  
variables that were declared to be in the cache and BOOL).  Though it  
doesn't look too hard to do.  All cached variables seem to be added  
through cmCacheManager::AddCacheEntry:

	http://public.kitware.com/cgi-bin/viewcvs.cgi/Source/cmCacheManager.cxx?revision=1.112&root=CMake&view=markup

So a quick hack would be to change the start of that function to:

	CacheEntry& e = this->Cache[key];
   	e.Properties.SetCMakeInstance(this->CMakeInstance);
	if ( value )
		{
		if ( type == cmCacheManager::BOOL )
			e.Value = cmSystemTools::IsOn(value) ? "ON" : "OFF";
		else
			e.Value = value;
		e.Initialized = true;
		}
	else
		...

But you're still stuck with having to #define ON and OFF... which I  
don't particularly care for.  So perhaps a parameter to configure_file  
would help?  It could tell CMake to turn '#cmakedefine  
VAR_THAT_IS_OFF' into  '#define VAR_THAT_IS_OFF 0'.  Or rather than a  
parameter, there might just a new keyword (such as '#cmakedefine01  
VAR_THAT_IS_OFF'.)

Any thoughts?

---Brian