Bug#787189: SDL_stretch.h causes compile error

Sun May 31 10:44:34 UTC 2015

Forgot to include the bug report again...

2015-05-31 11:43 GMT+01:00 Manuel A. Fernandez Montecelo
<manuel.montezelo at gmail.com>:
> 2015-05-31 7:14 GMT+01:00 Steve Clark <fawtytoo at gmail.com>:
>> On Sun, 31 May 2015 00:20:11 +0100
>> "Manuel A. Fernandez Montecelo" <manuel.montezelo at gmail.com> wrote:
>>
>>> 2015-05-30 21:27 GMT+01:00 Steve Clark <fawtytoo at gmail.com>:
>>> > On Sat, 30 May 2015 10:14:31 +0100
>>> > "Manuel A. Fernandez Montecelo" <manuel.montezelo at gmail.com> wrote:
>>> >
>>> >> Hello,
>>> >>
>>> >> 2015-05-29 15:54 GMT+01:00 Steve Clark <fawtytoo at gmail.com>:
>>> >> > Package: libsdl-stretch-dev
>>> >> > Version: 0.3.1-4
>>> >> >
>>> >> > Including the SDL_stretch header file causes the following
>>> >> > compile error:
>>> >> > /usr/include/SDL_stretch/SDL_stretch.h:25:23: fatal error:
>>> >> > SDL_video.h: No such file or directory #include "SDL_video.h"
>>> >> >
>>> >> > This is because the SDL stretch header files are in a separate
>>> >> > directory.
>>> >> >
>>> >> > There are 2 possible solutions:
>>> >> > 1. The SDL stretch header files should be in /usr/include/SDL
>>> >> > with the other SDL files
>>> >> >
>>> >> > 2. Comment the offending line out in SDL_stretch.h
>>> >> >
>>> >> > Option 2 is tested as workable because the header file continues
>>> >> > with: #ifndef  _SDL_video_h
>>> >> > #include <SDL/SDL_video.h>
>>> >> > #endif
>>> >> > ... which means it will still work. But I think option 1 is
>>> >> > preferable.
>>> >> >
>>> >> > I don't know if this is an upstream bug. But as it stands, it
>>> >> > unusable unless that line is commented out.
>>> >>
>>> >> Doesn't it work if you pass "-I/usr/include/SDL" (the first letter
>>> >> is a capital 'i', not an L) to the compiler?
>>> >>
>>> >>
>>> >> Cheers.
>>> >
>>> > Yes it does. But I shouldn't have to do that.
>>>
>>> As far as I am aware, it's a common practice to have library headers
>>> installed in different subdirectories of /usr/include (or under
>>> /usr/local/include, or /opt, or different paths of the user's
>>> home...), and then when compiling a project that needs another library
>>> included, to pass the -I option to the compiler.
>>>
>>> SDL_stretch is not a part of the main SDL modules, it comes from a
>>> different upstream, and they decided to install it in a different path
>>> rather than in the same directory as other SDL modules.
>>>
>>> The Debian package does not modify the installation path coming from
>>> upstream, so probably the situation is the same in other
>>> distributions.  Even within Debian, this has been the behaviour for
>>> many years, probably more than a decade by now, so moving the files to
>>> another path is not a good solution, I think -- it will affect the
>>> other users.
>>>
>>> On the other hand, the following code looks a bit problematic, because
>>> as it is happening to you right now, if the first #include fails the
>>> whole compilation fails, it is a fatal error to include a file and the
>>> file not being there, so those four lines do not make much sense.
>>>
>>> #include "SDL_video.h"
>>> #ifndef  _SDL_video_h
>>> #include <SDL/SDL_video.h>
>>> #endif
>>>
>>>
>>> The las version is from 6 years ago and this project has not been
>>> adapted for SDL2, so it makes me wonder if the project is still alive
>>> and if they will consider to fix it and give some advice on what to do
>>> in this case.
>>>
>>> I cannot do that at the moment, I will try at some point in summer if
>>> I find the time, but if in the meantime you or anybody else get in
>>> contact with upstream, please drop a line here.
>>>
>>>
>>> Cheers.
>>
>> If the first include is commented out, it will work. I was kinda hoping
>> you'd be willing to do that.
>
> I don't know if other members of the team have a different opinion, so
> this is not set in stone, but the problem that I see with changing the
> current behaviour and why I am reluctant to do it is the following: if
> for any reason in some user's systems the version of SDL/SDL_video.h
> that they want to use along with sdl-stretch is not in /usr/include
> (or one of the automatically searched directories), and rightly or
> wrongly they are relying in the current behaviour that was there for
> years, and they solve it by passing
> "-I/home/user/special-SDL-version/", removing that line will make
> things fail for them (or compile against an undesired version in
> /usr/include/SDL/SDL_*, instead of the ones that they want).
>
> And in general, I think that it is common and expected (if this is not
> so, please somebody correct me), that the users of a library have to
> explicitly pass the location with "-I", because upstreams don't always
> think the same and some headers include "SDL/SDL*.h", others expect -I
> to be added and remove the "SDL/".  Neither of the behaviours is
> "right" or "wrong", I think, it just happens to not be standardised.
>
> For example we have this case with libsdl2-mixer because one of the
> dependencies changed their behaviour in the same respect, and we had
> to adapt:
>
>   https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=726768
>
>
>> I will check upstream too. If SDL_stretch is no longer maintained, I
>> will probably fork it myself.
>
> OK, thanks.
>
> If it is still useful for people, it would be nice to have it working
> with SDL2, and perhaps in more architectures than x86 (now ARM and
> others are becoming very popular).
>
>
> Cheers.
> --
> Manuel A. Fernandez Montecelo <manuel.montezelo at gmail.com>

-- 
Manuel A. Fernandez Montecelo <manuel.montezelo at gmail.com>