On 02/28/2012 06:24 PM, Paul Eggert wrote:
> On 02/28/2012 09:01 AM, Pádraig Brady wrote:
>
>> Doing that would be "dangerous" as I said above,
>> or in other words, surprising to users to not update
>> possibly non NUL data in the output file.
>
> It's not surprising at all. It's what I want and
> expect. For example, suppose every nonzero byte of file A
> has an offset that is in a hole of file B. Then I should
> be able to overlay A and B into another file C, as follows:
>
> dd if=A of=C conv=sparse
> dd if=B of=C conv=sparse,notrunc
>
> This is akin to the "or" operation on files, and it'd be a nice
> operation to have. Even if A's contents don't fit in B's holes,
> I still might want to do the above, to allow B's non-holes to
> override A's contents. Why disable this useful functionality?
>
> Any surprise issues can be dealt with by documenting dd's
> behavior appropriately.
Fair enough. So this is the "write mask" functionality
I referred to, and which is now supported consistently
since the write(fd,"\0",1) to ftruncate() change.
>> +Try to seek rather than write @sc{nul} output blocks.
>> +This will create sparse output when extending.
>> +This option is ignored in conjunction with
>> +@samp{conv=notrunc} or @samp{oflag=append}.
>
> I still dubious about this level of handholding.
> dd is meant for low-level use, and as far as possible
> options should be orthogonal. For example, with dd,
> oflag=append does not disable seek=N -- both flags operate,
> which means that the seek is ineffective unless it
> is past the end of file. conv=sparse oflag=append
> should be similar: all it should mean is that the
> writes are sparse when they're past the end of
> the file (this latter functionality should work, but
> doesn't work with the proposed patch).
OK I'll remove the 2 restrictions, and essentially
move the comments for them from code to texinfo.
cheers,
Pádraig.
