[CMake] include_directories(...) versus set_directory_properties(PROPERTIES INCLUDE_DIRECTORIES ...)

[Marcel Loose]

Hi all,

Should I file this as a bug in the issue tracker then?

Best regards,
Marcel Loose.


op 03-02-14 17:57, David Cole schreef:
> I think it's mostly just a speed/space trade-off. If you don't want to
> consume the RAM temporarily (and in this case, I think it's a given,
> that we really don't want to consume 14G of RAM.....) then we'll have to
> spend the time doing the de-duplication on the way in.
> 
> Although, the data structures may not be in the right places already,
> they could easily be added.
> 
> I can't remember, though: is there a reason why we wouldn't want or be
> able to expand an input to an absolute path at the time it's given
> (current directory issue, or something else?) -- because a prerequisite
> to de-duplicating is referring to the directories in their full path
> canonical fashion.
> 
> That's just the quick brain dump version. If necessary, I can actually
> look at old commits, or the current code to try to refresh my memory.
> 
> 
> D
> 
> 
> 
> -----Original Message-----
> From: Brad King <brad.king at kitware.com>
> To: Marcel Loose <loose at astron.nl>
> Cc: cmake <cmake at cmake.org>; David Cole <dlrdave at aol.com>; Stephen Kelly
> <steveire at gmail.com>
> Sent: Mon, Feb 3, 2014 11:51 am
> Subject: Re: [CMake] include_directories(...) versus
> set_directory_properties(PROPERTIES INCLUDE_DIRECTORIES ...)
> 
> 
> On 01/30/2014 03:08 AM, Marcel Loose wrote:
>> cmake was consuming a whopping 14GB(!) of RAM memory.
> [snip]
>> include_directories() is used inside a macro that is called
>> quite a lot in order to resolve cross-dependencies. This never was a
>> problem, because CMake automatically performed de-duplication of 
> include
>> paths; until version 2.8.8.
> 
> IIUC it still does de-duplication but delays doing so until the
> generation step.  Therefore in this use case the configuration
> step buffers the duplicates.
> 
> Steve, David, having worked on these changes, what do you think
> about teaching include_directories and target_include_directories
> to skip appending a path that already exists in each dir/target
> property?
> 
> -Brad
> 
> 
>