On Tue, Oct 09, 2018 at 08:09:32AM +0900, Junio C Hamano wrote: > Julia Lawall <julia.lawall@xxxxxxx> writes: > > >> Doing the same for -S is much harder at the machinery level, as it > >> performs its thing without internally running "diff" twice, but just > >> counts the number of occurrences of 'foo'---that is sufficient for > >> its intended use, and more efficient. > > > > There is still the question of whether the number of occurrences of foo > > decreases or increases. > > Hmph, taking the changes that makes the number of hits decrease > would catch a subset of "changes that removes 'foo' only---I am not > interested in the ones that adds 'foo'". It will avoid getting > confused by a change that moves an existing 'foo' to another place > in the same file (as the number of hits does not change), but at the > same time, it will miss a change that genuinely removes an existing > 'foo' and happens to add a 'foo' at a different place in the same > file that is unrelated to the original 'foo'. Depending on the > definition of "I am only interested in removed ones", that may or > may not be acceptable. I think that is the best we could do for "-S", though, which is inherently about counting hits. For "-G", we are literally grepping the diff. It does not seem unreasonable to add the ability to grep only "-" or "+" lines, and the interface for that should be pretty straightforward (a tri-state flag to look in remove, added, or both lines). -Peff
