[PolyORB-users] Case clash in idls/

[Martin Krischik]

Am Sonntag, 6. Februar 2005 20:39 schrieb Simon Wright:
> > As true with many cvs archives. As you know cvs isn't ClearCase or
> > SubVersi= on=20
> > and can't do rename and directory versions - and -Pd is the only
> > workaround= =20
> > available.
> >
> > BTW: did you have a look at my build-polyorb? - it has a "get" option
> > using= =20
> > "-C -P -d" - reconf overwrites a few automake files and without -C you
> > will= =20
> > never see new automake files.
>
> Don't quite follow this, if support/reconfig is going to overwrite them?

Unless you use -C cvs will preserve the overwritten versions. 

> At the moment I'm trying to get PO to build for Windows under MinGW,
> with very little in the way of success. This is with GNAT 5.02a1 &
> mingw 3.2.0 rc.
>
> * idlac calls the CXXCPP with a strange mixture of slashes;
>
>     .\../cos/naming/CosNaming.idl
>
>   for example. This works fine unless the idl has #includes, when the
>   mingw native gcc fails. The only way I could get this to work was
>
>     CXXCPP='c:/GNATPRO/5.02a1/bin/gcc -E' ../polyorb/configure
>
> * support/adacompiler has race condition, need to put 'sleep 1' before
>   'mv $newname $1' at end or it doesn't find the new .o
>
> * links with native-linker (eg po_catref) fail because gnatlink can't
>   find the native linker (it says); remove the
>   --LINK=../../support/native-linker and it's OK
>
> * links with target-linker (eg po_cos_naming) eat all memory & die, I
>   just touched the .exe because I can use the omni name service or a
>   Linux polyorb one (I could try removing --LINK= and changing all the
>   .la's to .a)
>
> and after all that I can't get my test software to actually work. At
> least that part of it may be to do with running Windows under VMWare
> here!

Gosh, you got far more problems with MinGW than I did. The last I used Windows 
it failed on the 8192 character limit on the command line. And there is not 
much you can to about that. In the end I stopped making a Windows compile and 
"net use"d the Linux installation using "-aI" instead of "-I" for Windows. I 
take it you know what that means...

I don't consider it a polyorb problem. I think it is a ld/ar problem. Using 
the commandline to pass the module names for a static/share lib is not very 
scaleable. Even on Linix where the commandline is 32k in size.

The only possible fix would be to break down the link process using temporary 
ar archives. But that would be a significant amount of work deep inside the 
automake process. But I think it is doable.

> I'll give your cleanup script a try, though it takes so long for a
> build under MinGW that I'm losing the will to live ...

True - the MinGW compile is just SLOW. BTW: my script is optimized for single 
download / multi platform. I download once in Linux and "net use" the 
download inside a VMWare session.

Martin 

-- 
Martin Krischik
mailto://krischik@users.sourceforge.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : /pipermail/polyorb-users/attachments/20050207/c97902af/attachment.bin