MATLAB Answers

alelap83
0

Regexp lookbehind and lineanchors

Asked by alelap83
on 12 Sep 2019
Latest activity Edited by alelap83
on 16 Sep 2019
Could someone help me to understand why
st = ' a b c';
pattern = '(?<=^\s*)c';
regexp(st,pattern,'lineanchors')
ans =
[]
i.e., does not match (as I expected), while
st2 = [newline,st];
regexp(st2,pattern,'lineanchors')
ans =
7
i.e., finds a match?
My intent is to match 'c' that is preceded by the beginning of a line and zero or more white character. How should I do?

  2 Comments

Getting an output of 7 seems like a bug to me. Strangely the bug occurs even if the "zero or more matches" character does not even exist in the input string (R2012b):
>> regexp([char(10),st],'(?<=^_*)c','lineanchors') % Underscore is not in st.
ans =
7
>> regexp([char(10),st],'(?<=^)c','lineanchors') % expected
ans =
[]
>> regexp(st,'(?<=^_*)c','lineanchors') % expected
ans =
[]
What MATLAB version are you using?
You should report this as a bug, giving a link to this thread.
R2019a. Reported to Technical Support.
Edit: bug confirmed. Excerpt from Matlab Support's answer:
Indeed the behavior that you observed is indeed a bug in "regexp", which the developers are now aware of, and which might be addressed in some future release.
However, a workaround does exists, which consists in giving up on using the 'lineanchors' option (which makes the "^" and "$" metacharacters match embedded newlines too), and rely on grouping the (absolute) beginning of line "^" and the embedded newline "\n" as two alternatives.

Sign in to comment.

1 Answer

Answer by per isakson
on 13 Sep 2019
Edited by per isakson
on 16 Sep 2019
 Accepted Answer

"My intent is to match 'c' that is preceded by the beginning of a line and zero or more white character."
In the character array, ' a b c', the character, 'c', is (after the beginning of the line) preceded not only by whitespace but also by the characters 'a' and 'b'. Thus, [] is the expected result. Try
%%
chr = ' a b c';
xpr = '(?<=^[ ab]*)c';
regexp( chr, xpr, 'match', 'lineanchors' )
that returns
ans =
1×1 cell array
{'c'}
I fail to understand the behavior of your second example. I expect [], not 7. It's looks like a bug to me.
/R2018b
ADDENDUM
I learned something about the option,'once', the other day. It affects the type of the output. In this case the output is a character row instead of a cell array containing the character row. Thus,
>> regexp( chr, xpr, 'match', 'lineanchors', 'once' )
ans =
'c'

  2 Comments

Thank you. It works. Unfortunately, my example is a simplification of my real scenario where a b c are more complicated expressions and I cannot use this method.
"Could someone help me to understand why" I think I did that.
I cannot help regarding the "real scenario" because of lack of information.

Sign in to comment.