Re: [Wrapper-user] Single wrapper release possible?

[Leif M. <le...@ta...>]

Mike,
That is actually exactly how I have been waning to eventually do this. I
have
gotten support checked into CVS for naming the native library based on the
platform and architecture.

On Windows, the file can be named any of the following:
wrapper-windows-x86.dll
wrapper-windows.dll
wrapper.dll

On Linux, the following:
libwrapper-linux-i386.so
libwrapper-linux.so
libwrapper.so

On Solaris, the following:
libwrapper-sunos-sparc.so
libwrapper-sunos.so
libwrapper.so

I need to get the possible names for the other supported platforms resolved
by working with the volunteers who help with those builds.

The next step is to come up with a shell script which is capable of
doing the
same thing with the wrapper binary. Once that is done then a full cross
platform release will be possible.

Cheers,
Leif

Mike Castle wrote:

>In article <106...@we...>,
>Fredrik Israelsson  <wra...@li...> wrote:
>  
>
>>I don't question writing the wrapper in C. My question is more like why one 
>>cannot make one wrapper release with binaries needed for all platforms. I mean, 
>>storage space for the few MB:s needed is not a problem theese days.
>>    
>>
>
>I'm of two minds about this myself.
>
>After all, the JRE doesn't come with only one package; you have to download
>a different package for each arch.  Ok, well, Sun's is like that; I have no
>experience with any others, so maybe they do.
>
>But on the other hand, personally I package my app separately with all
>wrapper stuff appropriate for supports platforms, and distribute all addons
>(JDK, 3rd party libs, etc) as a second package, which is platform specific.
>
>Now, currently we only support Linux and Win32, so there are no conflicts
>with wrapper and wrapper.exe and wrapper.dll and libwrapper.so.
>
>But, what if I want to package another unix like platform that also uses
>.so?  Then it becomes a bit trickier.
>
>Sure, I could have wrapper.Linux and wrapper.FreeBSD and modify wrapper.sh
>to call wrapper.`uname -s` or something.  But that doesn't solve the
>dynamically loaded library issue.
>
>I think a nice enhancement would be if the call to System.loadLibrary()
>first tried "wrapper" and if that didn't work, try "wrapper.arch" or
>something.  I'm no Java expert, but it looks like the java property os.name
>might be appropriate?  Actually, may be a good idea to use os.name +
>os.arch to correctly support same OS on multiple platforms.  Linux and
>FreeBSD, for example, both work on non-i386 arch.  And if you're supporting
>FreeBSD, which has to build their own java from Sun's source anyway, might
>as well support them compiling it on a myriad of platforms.
>
>One reason this really concerns me is that the developers here like to run
>the same compiled code on two machines to test communications.  Sometimes
>those two machines are different OSes.  We're lucky now that Linux and
>Win32 can coexist.  But if we start supporting other platforms, it becomes
>more of an issue.
>
>Anyway, just my thoughts on the matter.
>
>Cheers,
>mrc
>  
>