Hi Ceriel, Quoting [Ceriel Jacobs] (Jan 24 2008):
Hi,
I'm really, really sorry ...
You better are!! ;-)
I have some more remarks now on the iovec specifications and the readv and writev specifications: this all concerns the "offset" field.
- the set_offset method does not consider the current value of len_in. Is it not an error if size >= 0 && len_in + offset > size? Maybe only if the buffer is not implementation-managed? - the set_len_in method does not consider the current value of offset. Is it not an error if size >= 0 && len_in + offset > size? Maybe only if the buffer is not implementation-managed? - the readv notes mention that if no len_in is specified, the buffer size is used as len_in. But what if offset > 0? - the above remark also holds for writev.
Yes - spec mostly ignores out-of-bound errors.
As far as I am concerned, the "offset" field can completely be removed from the specs. POSIX iovec does not have an offset field, and I don't know why it made it to the SAGA specs.
To simplify the reuse of memory segments w/o the need for pointer arythmetics: char mem[1024]; saga::iovec iov_1 (mem, 1024, 0, 512); saga::iovec iov_2 (mem, 1024, 512, 512); std::vector <saga::iovec> iovecs; iovecs.push_back (iov_1); iovecs.push_back (iov_2); file.readv (iovecs); without offset that woyuld work as well, as: char mem[1024]; saga::iovec iov_1 (mem , 512, 512); saga::iovec iov_2 (mem + 512, 512, 512); std::vector <saga::iovec> iovecs; iovecs.push_back (iov_1); iovecs.push_back (iov_2); file.readv (iovecs); although size and len_in are somehwat redundant here. If offsets are really needed can certainly be disputed. In C++, where pointer arithmetic is easy, one could certainly do without. I'd actually thought that it helps in Java. Ah, well... ;-)
Cheers, Ceriel
P.S. Sorry that I keep on having remarks, but these are things that come up while actually implementing it. Apparently, just reading the specs is not good enough.
Yep, I guess so. We are grateful for the comments: much better to have fixes and errata than a faulty spec! Thanks! Cheers, Andre.
Thilo Kielmann wrote:
Thank you, Ceriel! ;-)
Thilo
On Thu, Jan 24, 2008 at 03:42:14PM +0100, Andre Merzky wrote:
From: Andre Merzky <andre@merzky.net> To: Thilo Kielmann <kielmann@cs.vu.nl> Cc: saga-rg@ogf.org Subject: Re: [SAGA-RG] Fwd (ceriel@cs.vu.nl): [saga-cvs] "SAGA-RG/SAGA-Bindings/Java/OGF/src/org/ogf/saga/file FileFactory.java"
Yep, thats an error - all c'tors should be able to throe NotImplemented.
So, thats a good occasion to test the errata procedure :-)
Thanks, Andre.
Quoting [Thilo Kielmann] (Jan 24 2008):
From: Thilo Kielmann <kielmann@cs.vu.nl> To: saga-rg@ogf.org Subject: [SAGA-RG] Fwd (ceriel@cs.vu.nl): [saga-cvs] "SAGA-RG/SAGA-Bindings/Java/OGF/src/org/ogf/saga/file FileFactory.java"
Hmm,
do we now have the first item for the "errata" document? ;-)
Thilo
----- Forwarded message from Ceriel Jacobs <ceriel@cs.vu.nl> -----
To: saga-cvs@cct.lsu.edu From: Ceriel Jacobs <ceriel@cs.vu.nl> Subject: [saga-cvs] "SAGA-RG/SAGA-Bindings/Java/OGF/src/org/ogf/saga/file FileFactory.java"
Update of /projects/SAGA-RG/SAGA-Bindings/Java/OGF/src/org/ogf/saga/file In directory asylum.cct.lsu.edu:/tmp/cvs-serv30810/src/org/ogf/saga/file
Modified Files: FileFactory.java Log Message: Inconsistency in the SAGA specification: Buffer constructor throws NotImplemented, but IOVec does'nt? In the Java version, it does now!
_______________________________________________ saga-cvs mailing list saga-cvs@cct.lsu.edu https://mail.cct.lsu.edu/mailman/listinfo/saga-cvs ----- End forwarded message ----- -- No trees were destroyed in the sending of this message, however, a significant number of electrons were terribly inconvenienced. -- No trees were destroyed in the sending of this message, however, a significant number of electrons were terribly inconvenienced.