Conditional compilation based on GSL version

Conditional compilation based on GSL version

[M Joonas Pihlaja]

Hi,

Could GSL expose some facility in gsl_version.h to compare the version 
of GSL at compile time for dumb clients which don't want to impose 
special build system requirements?  The current GSL_VERSION define 
exports a string like "1.12", but that's not so useful for testing 
against versions using the C preprocessor at compile time.  Something 
simple like GSL_VERSION_MAJOR and GSL_VERSION_MINOR defines would go a 
long way.  Or perhaps a more complex macro like 
GSL_VERSION_IS_COMPATIBLE(major,minor) would be preferable.

My current workaround is to grab the GSL version string using 
pkg-config, split it into major and minor version numbers, and pass 
those in as preprocessor defines.  It's rather ugly, possibly error 
prone, and of course there's also the pkg-config requirement, so it's 
not really ideal.  I did look in the manual, but if I'm missing the 
obvious solution please whack me with your stout clue stick.  :)

Cheers,

Joonas

Re: Conditional compilation based on GSL version

[Brian Gough]

At Thu, 19 Feb 2009 23:04:40 +0200 (EET),
M Joonas Pihlaja wrote:
> Could GSL expose some facility in gsl_version.h to compare the version 
> of GSL at compile time for dumb clients which don't want to impose 
> special build system requirements?  The current GSL_VERSION define 
> exports a string like "1.12", but that's not so useful for testing 
> against versions using the C preprocessor at compile time.  Something 
> simple like GSL_VERSION_MAJOR and GSL_VERSION_MINOR defines would go a 
> long way.  Or perhaps a more complex macro like 
> GSL_VERSION_IS_COMPATIBLE(major,minor) would be preferable.

I'd say the GNU approach to compatibility is to test for the presence
of individual functions or features with autoconf, rather than package
versions--it is more reliable in the long-term.  So if you can use
autoconf, I would recommend that.  If not, maybe you could describe
the details of the situation a bit more.

-- 
Brian Gough

Re: Conditional compilation based on GSL version

[M Joonas Pihlaja]

On Fri, 20 Feb 2009, Brian Gough wrote:

> At Thu, 19 Feb 2009 23:04:40 +0200 (EET),
> M Joonas Pihlaja wrote:
> > Could GSL expose some facility in gsl_version.h to compare the version 
> > of GSL at compile time for dumb clients which don't want to impose 
> > special build system requirements?  

> I'd say the GNU approach to compatibility is to test for the 
> presence of individual functions or features with autoconf, rather 
> than package versions--it is more reliable in the long-term.  

Where to draw the line though?  Should I check for the availability of 
every single symbol I'm using from GSL?  Just the ones which I know 
are volatile or new?  *shudder*  Versioning the API is a perfectly 
valid solution to the problem and since GSL does have API stability, 
why not make the version conveniently available to client programs as 
well?

> So if you can use autoconf, I would recommend that.  If not, maybe 
> you could describe the details of the situation a bit more.

Of course.  I won't go into details on why I don't use autoconf for my 
code, but suffice to say that it's about as portable as a pile of 
bricks; doable with enough effort, but really inconvenient and not fun 
at all.  #ifs work everywhere and have a minimal burden when kept 
in check.

The specific situation I'm looking at is simply that GSL implemented 
missing complex vector operations in 1.12 which I'd like to export in 
my GSL <-> Lua minibinding.  Autoconf doesn't even know what Lua is, 
never mind what to do with it. :)

Cheers,

Joonas

Re: Conditional compilation based on GSL version

[Ed Smith-Rowland]

M Joonas Pihlaja wrote:
> On Fri, 20 Feb 2009, Brian Gough wrote:
>
>   
>> At Thu, 19 Feb 2009 23:04:40 +0200 (EET),
>> M Joonas Pihlaja wrote:
>>     
>>> Could GSL expose some facility in gsl_version.h to compare the version 
>>> of GSL at compile time for dumb clients which don't want to impose 
>>> special build system requirements?  
>>>       
>
>   
>> I'd say the GNU approach to compatibility is to test for the 
>> presence of individual functions or features with autoconf, rather 
>> than package versions--it is more reliable in the long-term.  
>>     
>
> Where to draw the line though?  Should I check for the availability of 
> every single symbol I'm using from GSL?  Just the ones which I know 
> are volatile or new?  *shudder*  Versioning the API is a perfectly 
> valid solution to the problem and since GSL does have API stability, 
> why not make the version conveniently available to client programs as 
> well?
>
>   
>> So if you can use autoconf, I would recommend that.  If not, maybe 
>> you could describe the details of the situation a bit more.
>>     
>
> Of course.  I won't go into details on why I don't use autoconf for my 
> code, but suffice to say that it's about as portable as a pile of 
> bricks; doable with enough effort, but really inconvenient and not fun 
> at all.  #ifs work everywhere and have a minimal burden when kept 
> in check.
>
> The specific situation I'm looking at is simply that GSL implemented 
> missing complex vector operations in 1.12 which I'd like to export in 
> my GSL <-> Lua minibinding.  Autoconf doesn't even know what Lua is, 
> never mind what to do with it. :)
>
> Cheers,
>
> Joonas
>
>   
Autoconf might not know what Lua is but CMake might.
I played around with CMake ans it seemed easier than Autoconf and it was 
loaded with pre-built things for recognizing tons of stuff,

Good Luck,

Ed

Re: Conditional compilation based on GSL version

[Brian Gough]

At Fri, 20 Feb 2009 14:09:40 +0200 (EET),
M Joonas Pihlaja wrote:
> The specific situation I'm looking at is simply that GSL implemented 
> missing complex vector operations in 1.12 which I'd like to export in 
> my GSL <-> Lua minibinding.  Autoconf doesn't even know what Lua is, 
> never mind what to do with it. :)

Ah, wrapping the whole library is a different case from what I was
thinking of--an application using some specific functions.

It could be useful to have those version numbers in the header but as
they aren't in any of the old ones I guess it is not really worth
adding them now.

Expanding GSL_VERSION using the C preprocessor on the gsl header file
is one way that doesn't require those package config scripts.

-- 
Brian Gough