Re: [Bug-ddrescue] bug on non-hardbs rounded start position or size!

[Antonio Diaz Diaz]

Hi Florian,

Florian Sedivy wrote:
> To stay true to domain limits, ddrescue could issue a corrected hrdbs
> rounded read in direct mode, but only use the part that was
> requested.

The problem with this is that, in addition to throwing away successfully 
read data, it would slow ddrescue for everybody. Each write would have 
to be checked for bounds.

In the spirit of ddrescue of never reading a good sector twice, I think 
that not attempting partial sector reads in direct mode is a better 
solution.


> This solution could even be automatic, as the only negative effect
> would be a potentially unnecessary second read of the same block,
> when the other halve is eventually requested in another run. Yes, it
> means throwing away successfully read data, but this is generally in
> the responsibility of a user requesting a non-rounded domain, and no
> different from what really happens when reading the same domain from
> a block device without direct mode.

The difference is, I think, that without direct mode the other half may 
be read from the kernel cache. But in direct mode, reading the other 
half means a second disc access.


Best regards,
Antonio.