On Sun, Jul 5, 2015 at 8:32 PM, Jim Meyering <j...@meyering.net> wrote: > Here is a patch to teach grep the --files0-from=F option. > That is the same option that is already supported by wc, du and sort > in the GNU coreutils. > > This option allows one to specify NUL-separated file names in the file > argument, F. When I would receive requests to add this option to other > programs from the coreutils, I would refuse, because one can obtain > similar behavior by using xargs like this: > > xargs -0 PROGRAM < F > > The "use xargs" argument does not work for programs like wc, du and sort. > In those cases, concatenating the output from multiple invocations > (as that xargs usage would do) does not result in the same output as > using --files0-from=F, e.g., because of totals or, for sort, because > the result cannot be the concatenation of separately-sorted outputs. > > For most uses of grep, it would be ok to process separate lists of > file names in batches and to concatenate the results, but not with > the --count (-c) option. The count must reflect the total over > all files processed.
Hmm... *this* --count option does not have that behavior, so it's not a good justification. This is a per-file and per-output-line count. > Also, more subtly, consider what would happen when the last batch of file > name inputs to xargs happens to contain only one name. The output from > that final invocation of grep (assuming no option like -h or -H) would > omit the "FILENAME:" prefix for any lines matched from that final file. And in this case, simply adding '/dev/null' or using -h ensures that even that unusual case will be handled properly. Now that I've removed those bits of justification, I'll have to think more about whether it's worth adding this.
