[bug #34020] RFE: Improve handling of sparse files

[Paolo Bonzini]

Follow-up Comment #1, bug #34020 (project grep):

This is not a bug.  The problem is not the sparseness of the file.  The
problem is that grep works on lines, and lines are separated by LF characters.
 So, a file that has 2GB of NULLs still has a single line.

This bug in particular is not fixable when grep's input is a pipe (so you
cannot rewind the input when printing).

grep used to have some old code that used mmap to read the file.  We ripped
that out because in normal cases it actually had worse performance than
read(), but in this case it had an advantage: the kernel has the possibility
to swap out (and back in) to the sparse file.  It would still require 100% CPU
and can fail on 32-bit machines due to the small address space.

However, care must be taken for code like this:

   echo foo > bar
   (grep foo; cat) < bar

Here, the "cat" command must output nothing.  IIUC the old mmap code did not
advance the file pointer, so that the "cat" would print again the whole
contents of the input file.

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?34020>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/