🚀 go-pugleaf

We're investigating a weird issue that has cropped up.

We have software that transfer data from device to device and we're
finding that when the target device is a file on a Windows machine a
random number of bytes are not being transferred at the end.

The software is not new and is based on technology that is over a
decade old. Put another way:Until this morning this was a tried and
trusted utility that is being used dozens of times a day. It is vital
to part of our business and thankfully to our older NetWare servers is
working just fine.

The basic application consists of a loop reading from a source handling
object and writing to a target handling object ie;

while( not_done )
{
  source->read_chunk(); // Chunk <=64kB
  target->write_chunk();
}

The application generates an MD5 checksum as it transfers and since
this is the same in all cases it tells us that we are reading
everything we ought to (we have generated test data and verified the
actual MD5).

Furthermore the target creates a control file at the end of the process
and this records the highest block it was asked to write out. This
value is what it ought to be.

The only thing wrong is that the actual output file is shorter by
anything up to a couple of megabytes.

Our transfer is performed in no more than 64kB chunks.

The Windows machines (Win2K) report no problems in their log file.

I vaguely recall someone (Alan Bellingham?) reporting an issue with
file I/O and data getting lost. Does this ring any bells?

-- 
Andrue Cope [TeamB]
[Bicester, Uk]
http://info.borland.com/newsgroups/guide.html