Are answers comparing existing answers acceptable?

Question

A "not an answer" flag was raised against this answer. As you can see, it is not actually providing any new solutions, just comparing the ones already provided by other answers and showing which one is faster.

On the one hand, it is not an answer. It is not attempting to solve the issue. It is, however, providing useful information that can help users choose between the proposed methods.

Personally, I lean towards allowing that type of post. I have to, I have been guilty of posting such myself. In my own answer, I simply collected all posted solutions and compared them in terms of speed. That answer has received 9 upvotes to date and no downvotes, which (might) suggest that the community feels that such comparisons are useful.

So, how so we feel about comparative answers? Answers that provide no new method of solving the OP's issue but do provide useful information about the existing answers. Do we like them? Do we not? Should they be deleted? Or do we feel that they add useful information and can be tolerated?

Answer 1

I observe two things about that answer:

1. Even if you've read every other answer, that one still offers additional information regarding the solutions proposed elsewhere.

2. Even if every other answer was deleted, that answer would still allow a reader to solve the problem presented in the question.

Therefore, it isn't completely redundant and it can stand alone.

Quick update, since a few people seem to have assumed I was planning on deleting the other answers or something: observation #1 establishes that the answer isn't redundant, while observation #2 establishes that the answer is actually an answer - that while it builds on the work done by other answers, it does so without becoming dependent on them.

At worst, I see no harm in it, and for some readers it may offer significant benefits.

Answer 2

Does it answer the question? Clearly; in fact, it gives four solutions. That alone makes it an answer (i.e. not not-an-answer).

So, the question is, should it be downvoted?

Does it add value compared to the other solutions? Yes, it gives objective benchmarks. Those can be very helpful in figuring out which solution to use for in and similar situations.

If anything should be changed about this post, it's that it should give credit to the authors of the posts from which his solutions are taken.

Answer 3

I think that enhancements or clarifications to existing solutions are Ok, as long as long as the resulting post is itself an answer. Case in point, some years ago I added the following answer: “unpacking” a tuple to call a matching function pointer. This formally adds absolutely nothing new to the answer by Johanne, but it has 24 upvotes and no downvotes at this time of writing, so presumably people find it useful.

Answer 4

Are answers comparing existing answers acceptable?

No. Those answers are essentially regurgitating someone else words. They do not give anything novel back (frankly, I don't trust any Joe Doe performance tests, not even mines), and just serve to clutter the answers further with the information already available somewhere else.

If someone considers that one solution is fastest/cheaper/awesome they can readily express them by upvotes and/or comments complementing the information on the answer.

I agree that the NAA flag wasn't exactly the flag it should be raised here, but a "other" -> "this answer doesn't add anything new and simply repeats the solutions of other answers", which I believe any moderator would acted by deleting the answer.

So, when comparing existing answers is a valid answer? When your very own solution is part of the comparison:

I recommend fooring the bar with X tool, since the times are 20, 23 and 40 percent faster than Joe, Marc, Jhon solutions.

Otherwise, you are just cluttering the answers. If you want to add your own comparison, just comment on the specifics answers you benchmarked (or whatever variable you were comparing).

Answer 5

The original question is already a duplicate and now we have more duplicated information, that is, in addition, not really critical of the solutions used (e.g. using the default grep instead of grep -F which can be used as there are no patterns involved (muru's answer).

IMO it is preferable to have this answer in a separate Q&A pair, whereby the title of that question should be adapted to what the answer answers, i.e. includes information about the speed comparison. That way someone might actually hit upon the answer, now googling for text processing speed comparison unix.stackexchange.com doesn't seem to show things (but that might be my googling deficiency). A link to the original question and answers with appropriate credits would of course be necessary.

Answer 6

I really like comparisons of different tools. When we decide which tool should be used for the given job then time needed for particular operation is one of the most important thing (the others are availability, portability, etc). So yes, for me it is perfectly acceptable answer, and at least once I gave similar comparison: https://unix.stackexchange.com/a/156602/80886.

---

Now, with that being said I have to admit that discussed benchmark looks fishy to me because I can hardly believe that 3 consecutive grep are 6x faster than single grep, and 10x faster than single awk. Really???

I made a fast xcheck:

# creating the file by repeating next 'cat' until abc.txt=196MB
$ cat /usr/src/linux/Documentation/**/*.txt >> abc.txt

Results:

• single awk:

  $ time bash -c "awk '/a/ && /b/ && /c/' abc.txt >> /dev/null"
  7.75s user 0.23s system 99% cpu 7.984 total
  

• triple grep:

  $ time bash -c "grep a abc.txt | grep b | grep c >> /dev/null"
  5.65s user 1.41s system 188% cpu 3.745 total
  

• single grep

  $  time bash -c "grep -e 'a.*b.*c' -e 'a.*c.*b' -e 'b.*a.*c' -e 'b.*c.*a' -e 'c.*a.*b' -e 'c.*b.*a' abc.txt >> /dev/null"
  2.43s user 0.13s system 99% cpu 2.565 total
  

• single perl:

  $ time bash -c "perl -ne 'print if /a/ && /b/ && /c/' abc.txt >> /dev/null"
  11.95s user 0.21s system 99% cpu 12.174 total
  

I'm not saying that something is incorrect in that answer, but above results make more sense to me:

• grep as a dedicated tool is fastest because it is designed and optimized to do only one job, and do it fast.
• single grep is faster than 3, because no time is wasted for forking
• perl is slowest as it is the largest program

---

_{Test done on weak netbook with atom processor, so only relative times make sense.}