Fix -w behaviour and docs (<https://github.com/petdance/ack2/issues/445>)

[petdance]

Thanks for the patch. Something of this magnitude needs some severe test coverage before I can even think of merging it in.

[epa]

I will add some tests.

[petdance]

I'm also concerned about any potential speed degradations.

The change you're suggesting here needs to be discussed on the ack-users mailing list because it is potentially a huge breakage of existing behavior. The new -w may be "fixed", but it is changing existing behavior and we have to be very very very careful with that.

[epa]

Would you prefer that I first document and test the existing behaviour, and then we can decide whether to change it?

[petdance]

I think that makes a lot of sense. If anything is going to be changed, then there will have to be a clear explanation of that change in the Changes file for any future users.

[epa]

I've sent a pull request which documents and tests the current semantics.

[epa]

I've posted on ack-user asking about this proposed change to -w semantics.

[epa]

There haven't been any comments on ack-users. What is your opinion on the proposed change?

[petdance]

https://groups.google.com/forum/#!topic/ack-users/HVKAvkO4G3U

[petdance]

I've only skimmed it. First thing that comes to mind is that we can't change anything related to parentheses until we get the highlighting bug fixed.

[epa]

I don't think this is really 'related to parentheses'; the parens are just one example test case that demonstrates the bug (where -w fails to require a whole word match). Here is another test case:

mkdir test; cd test
for i in box ox oxen; do echo $i >$i; done
ack -w '\w\w'

This should match a two-letter whole word, but it ends up matching all three strings. Or indeed

ack -w '[oa]x'

should match the whole words 'ox' and 'ax', but finds the not-whole-word match inside 'box' too.

[epa]

Do you have time for a closer look at the fix? As I mentioned it is not related to parentheses or any other particular regexp feature; it fixes the problem that if the given regexp doesn't begin with a word character then the -w flag fails to work. https://github.com/petdance/ack2/issues/14 is an older bug covering the same issues. Contributor hoelzro reviewed an earlier version of the patch in https://github.com/petdance/ack2/issues/445

[hoelzro]

@petdance What are your thoughts on this moving forward? No one on ack-users seems to mind that this would change.

[petdance]

How does grep handle this? If you grep -w mu., what behavior do you get?

Test cases to consider:

grep -w mu
grep -w mu.
grep -w mu.*
grep -w mu.+
grep -w .mu.
grep -w (mu)

[hoelzro]

$ echo must | ack -w mu.
must
$ echo must | grep -w mu.
(no output)

[epa]

As far as I know grep handles it correctly, in that the -w causes it to match whole words (and doesn't depend on whether the regexp given begins or ends with a word character).

% mkdir test
% cd test
% echo mu >>t
% echo mux >>t
% echo muxy >>t
% grep -w mu. t
mux

[petdance]

@epa: We need to stick to facts and not use words like "correct" or "wrong", because that's an opinion, and everyone has an opinion on things but they can get confusing.

[epa]

I do agree, which is why I clarified exactly what I meant by 'correct' in this case.

[epa]

Somewhat related: https://rt.perl.org/Public/Bug/Display.html?id=127670

[petdance]

The behavior of -w is not going to change in ack2. It is updated in ack3.