Re: [cp-patches] FYI: FileChannelImpl.c fixes

[Roman Kennke]

Hi Mark,

Am Montag, den 23.01.2006, 15:13 +0100 schrieb Mark Wielaard:
> Hi Roman,
> 
> On Tue, 2006-01-17 at 16:36 +0000, Roman Kennke wrote:
> > Here come some fixes for FileChannelImpl.c. Mainly this provides
> > alternative implementations of the methods for systems that don't have
> > a filesystems and replace some ssize_t variables with jint for a little
> > more correctness. Also this reimplements lock and unlock to use the
> > corresponding target native macros.
> > 
> > 2006-01-17  Roman Kennke  <address@hidden>
> > 
> >     * native/jni/java-nio/gnu_java_nio_channels_FileChannelImpl.c:
> >     (Java_gnu_java_nio_channels_FileChannelImpl_init): Improved
> >     exception messages a little.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_open): Provided
> >     alternative implementation for systems without filesystems.
> >     Replaced snprintf with the corresponding target native macro.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_implCloseChannel):
> >     Only do something when we have a filesystem.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_available): Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_size): Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_implPosition): Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_seek):
> >     Only do something when we have a filesystem.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_implTruncate):
> >     Only do something when we have a filesystem.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_mapImpl): Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_read__):
> >     Replaced ssize_t variables with jint. Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_read___3BII):
> >     Replaced ssize_t variables with jint. Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_write__I):
> >     Replaced ssize_t variables with jint. Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_force):
> >     Only do something when we have a filesystem.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_write___3BII):
> >     Replaced ssize_t variables with jint. Provided
> >     alternative implementation for systems without filesystems.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_lock): Reimplemented
> >     to use the corresponding target native macro.
> >     (Java_gnu_java_nio_channels_FileChannelImpl_unlock): Reimplemented
> >     to use the corresponding target native macro.
> 
> This broke File locking. In particular the mauve test
> gnu.testlet.java.nio.channels.FileChannel.lock and the eclipse osgi
> locking. How did you test this patch and on what platforms?
> 
> I have a hard time seeing how this patch is supposed to work. You
> replace the calls with new target layer functionality, but that
> functionality seems to depend on having things like
> TARGET_NATIVE_FILE_LOCKMODE_READ, which depend on things like
> HAVE_F_RDLCK being defined, but I cannot find where you exactly do this.

I tested this with the JamaicaVM. Sorry, I didn't run a full test with
JamVM/Classpath or similar. The HAVE_F_RDLCK thing very much sounds like
an autoconf test. I'll check this with Torsten and come back to it.

Roman

signature.asc
Description: Dies ist ein digital signierter Nachrichtenteil