Re: Sector file format (was: Re: Testing framework)

[Håkon Løvdal]

On 11/15/05, Patrick Leslie Polzer <address@hidden> wrote:
> On Tue, 15 Nov 2005 03:12:02 +0100
> Håkon Løvdal <Håkon Løvdal <address@hidden>> wrote:
> > * Writing is not considered.
> I think this should be implemented for easy working with the gathered
> data.

Supporting writing means that the content of the sectors has
to be versioned and the reader has to become stateful,
i.e. the content of sector x is
after 0 writes "AAA...",
after 1 write "BBB...",
after 2 writes "CCC...",
and if there are more than 2 writes this is an error. Less than 2 writes
should also probably generate an error.

Supporting only reading on the other hand is very simple because then there
always is a one to one relation to the content and a given sector. In order to
support creating binary error-report-disk-snapshots only reading is required,
and this in my opinion the most important part to implement first.
But write support will of course be very useful, I am not opposing it :)


> > * For simplicity I use environmental variables to pass
> >   arguments to filespy.
> Are you planning to do in this way in the final version, too?
>
> > * I use a nice and simple (external) library called sl for storing a
> > list with the meta information, but it is very simple to install.
> Can't we use our own list implementation (or put 'sl' into our
> distribution)? That seems to be an unnecessary dependency to me.

Initially I wanted to write a completely general file access record/restore
mechanism and I compiled it as an object file overriding the
open/read/write/close functions (in various variants...) and used
LD_PRELOAD to load it. So that is why I started calling it filespy,
was using environmental variables and an external list library.

However I never got this to work reliably, and when this issue
was brought up again I thought that I at least should try to finish
what I had done so far and put it in as a prototype into parted.

BR Håkon Løvdal.