🚀 go-pugleaf

Well we progressed. After the success of FILE_FLAG_WRITE_THROUGH I
implemented code to test for file size not matching our expectation and
the problem has "gone away". "Absence of proof is not proof of absence"
and all that but tests suggest that:

  close file
  reopen file
  get file size

Ensures that no data is left off. It also doesn't matter whether
closing the file request a buffer flush first or not although we've
left that in.

I've done a lot of tests today and we do seem to achieved a fix. At
least we also know that if the fix isn't total we'll detect the issue
and be able to advise people to activate the write through method.

The moral of this seems to be that you can't trust Windows to write
data to a remote location. Coupled with the difficulty over what
provokes this (seems to be timing related - maybe even network traffic
related) and we have an uncertain resolution.

> (I cannot remember a disk which did not have 512 byte sectors other 
> than floppies which had power of two sub-multiples of that.)

Hard disks are normally 512 - but we've got that covered if we ever
need to look into it.

For now I'm tentatively assuming we've solved the problem. Either that
or we've made it extremely rare and detectable.
-- 
Andrue Cope [TeamB]
[Bicester, Uk]
http://info.borland.com/newsgroups/guide.html