13

Is there any reason not to use use --sparse=always with every invocation to cp?

info cp says:

‘--sparse=WHEN’
     A “sparse file” contains “holes”—a sequence of zero bytes that does
     not occupy any physical disk blocks; the ‘read’ system call reads
     these as zeros.  This can both save considerable disk space and
     increase speed, since many binary files contain lots of consecutive
     zero bytes.  By default, ‘cp’ detects holes in input source files
     via a crude heuristic and makes the corresponding output file
     sparse as well.  Only regular files may be sparse.

    The WHEN value can be one of the following:

...

    ‘always’
          For each sufficiently long sequence of zero bytes in the input
          file, attempt to create a corresponding hole in the output
          file, even if the input file does not appear to be sparse.
          This is useful when the input file resides on a file system
          that does not support sparse files (for example, ‘efs’ file
          systems in SGI IRIX 5.3 and earlier), but the output file is
          on a type of file system that does support them.  Holes may be
          created only in regular files, so if the destination file is
          of some other type, ‘cp’ does not even try to make it sparse.

It also says:

[...] with the following alias, ‘cp’ will use the minimum amount of space supported by the file system.

alias cp='cp --reflink=auto --sparse=always'

Why isn't --sparse=always the default?

13
  • 1
    It’s incompatible with --reflink, apart from that I don’t know... Sep 3, 2017 at 15:35
  • perhaps just because the developers wanted to utilise the principle of least surprise, or because POSIX specified otherwise? (is cp even in posix, i forget)
    – cat
    Sep 3, 2017 at 15:39
  • 2
    Checking for sparseness might be detrimal to performance, sparse files may cause severe filesystem fragmentation, and there was at least one instance of data corruption with cp --sparse. Sep 3, 2017 at 15:39
  • 1
    Copying the data for (mainly non-sparse) files through a read/write loop involves dma of data into and out of memory, whereas looking for runs of zeroes implied by always (or auto where the number of blocks doesn't match the file size) will drag the data into the cpu caches and involve much more cpu bandwidth and cycles.
    – meuh
    Sep 3, 2017 at 16:49
  • 1
    @StephenKitt It is compatible with --reflink: info cp contains: with the following alias, ‘cp’ will use the minimum amount of space supported by the file system. alias cp='cp --reflink=auto --sparse=always'
    – Tom Hale
    Sep 4, 2017 at 3:38

1 Answer 1

3

There are a few reasons why it is not default, one is backwards compatibility, performance, and last but not least, the principle of least surprise.

My understanding is that when you enable this option, there is CPU overhead which might not necessarily be acceptable, besides, backwards compatibility is also key. The cp command works reliably without, it does add a little space saving, but these days, that really is negligible, in most cases at least ...

I think the comments you received also highlighted other reasons.

Principle of least surprise means you do not change something needlessly, cp has been around for decades, changing its default behavior will upset many veterans.

1
  • How big is the CPU overhead of --sparse?
    – Atemu
    Dec 19, 2021 at 17:33

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .