Re: [Issue 1715] Checksums in the rA layer

["Peter N. Lundblad" <pe...@famlundblad.se>]

On Sun, 27 Jun 2004, Greg Hudson wrote:

> On Sun, 2004-06-27 at 04:52, Peter N. Lundblad wrote:
> > On Sat, 26 Jun 2004, Greg Hudson wrote:
> >
> > > On Sat, 2004-06-26 at 14:38, Peter N. Lundblad wrote:
> > > > - I'm sending text deltas and no checksums. Should I? (Same question also
> > > > applies to DAV. I will be consistent about this.)
> > >
> > > get_file has no checksums, so I'd say no, it's probably not worth it.
> > > Others might disagree, though.
> > >
> > Both the protocol file and the code look like get-file indeed has a
> > checksum. What do I miss?
>
> I guess it does.  I looked at svn_ra.h and saw no checksum.  (Which just
> means the ra_svn client code takes care of checking it, as it turns out,
> but I didn't think that far.)
>
> So, yes, I guess you want a checksum.
>
Hmmm... Studying it more carefully, I'm not sure again. For get_file, the
checksum check is internal to the protocols and not exposed to the client.
That works because the contents stream through the RA implementation in
fulltext. But in the case of get_file_revs, the RA implementation just
passes the delta to the client. Then, in the delta editors, the checksum
is also passed to the client.

Do you think it is useful to change get_file_revs to provide the checksum
to the caller? I mean, how necessary is the checksum in this case? We
expect the network traffic to be correct, don't we?

I think this is the only way to get checksum verification on this call.
Else we have to apply the text deltas once more just to get the fulltext.

I think I need advise here...

Thanks
//Peter

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [Issue 1715] Checksums in the rA layer

["Peter N. Lundblad" <pe...@famlundblad.se>]

On Sun, 4 Jul 2004, Greg Hudson wrote:

> On Sat, 2004-07-03 at 16:40, Peter N. Lundblad wrote:
> > Do you think it is useful to change get_file_revs to provide the checksum
> > to the caller? I mean, how necessary is the checksum in this case? We
> > expect the network traffic to be correct, don't we?
>
> I have no strong preference either way.
>
I skip checksums until someone objects:-)

> The purpose of our checksums is to prevent corruption, whether from a
> faulty filesystem or a faulty network.  But a corrupt "blame" result is
> not generally as harmful as a corrupt wc or repository, so it's not
> critical that we detect it.
>
Who knows what get_file_revs will be used for in the future? :-)

Regards,
//Peter

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [Issue 1715] Checksums in the rA layer

[Greg Hudson <gh...@MIT.EDU>]

On Sat, 2004-07-03 at 16:40, Peter N. Lundblad wrote:
> Do you think it is useful to change get_file_revs to provide the checksum
> to the caller? I mean, how necessary is the checksum in this case? We
> expect the network traffic to be correct, don't we?

I have no strong preference either way.

The purpose of our checksums is to prevent corruption, whether from a
faulty filesystem or a faulty network.  But a corrupt "blame" result is
not generally as harmful as a corrupt wc or repository, so it's not
critical that we detect it.


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org