How to get the inverse of a regular expression? - regex

Let's say I have a regular expression that works correctly to find all of the URLs in a text file:
(http://)([a-zA-Z0-9\/\.])*
If what I want is not the URLs but the inverse - all other text except the URLs - is there an easy modification to make to get this?

You could simply search and replace everything that matches the regular expression with an empty string, e.g. in Perl s/(http:\/\/)([a-zA-Z0-9\/\.])*//g
This would give you everything in the original text, except those substrings that match the regular expression.

If for some reason you need a regex-only solution, try this:
((?<=http://[a-zA-Z0-9\/\.#?/%]+(?=[^a-zA-Z0-9\/\.#?/%]))|\A(?!http://[a-zA-Z0-9\/\.#?/%])).+?((?=http://[a-zA-Z0-9\/\.#?/%])|\Z)
I expanded the set of of URL characters a little ([a-zA-Z0-9\/\.#?/%]) to include a few important ones, but this is by no means meant to be exact or exhaustive.
The regex is a bit of a monster, so I'll try to break it down:
(?<=http://[a-zA-Z0-9\/\.#?/%]+(?=[^a-zA-Z0-9\/\.#?/%])
The first potion matches the end of a URL. http://[a-zA-Z0-9\/\.#?/%]+ matches the URL itself, while (?=[^a-zA-Z0-9\/\.#?/%]) asserts that the URL must be followed by a non-URL character so that we are sure we are at the end. A lookahead is used so that the non-URL character is sought but not captured. The whole thing is wrapped in a lookbehind (?<=...) to look for it as the boundary of the match, again without capturing that portion.
We also want to match a non-URL at the beginning of the file. \A(?!http://[a-zA-Z0-9\/\.#?/%]) matches the beginning of the file (\A), followed by a negative lookahead to make sure there's not a URL lurking at the start of the file. (This URL check is simpler than the first one because we only need the beginning of the URL, not the whole thing.)
Both of those checks are put in parenthesis and OR'd together with the | character. After that, .+? matches the string we are trying to capture.
Then we come to ((?=http://[a-zA-Z0-9\/\.#?/%])|\Z). Here, we check for the beginning of a URL, once again with (?=http://[a-zA-Z0-9\/\.#?/%]). The end of the file is also a pretty good sign that we've reached the end of our match, so we should look for that, too, using \Z. Similarly to a first big group, we wrap it in parenthesis and OR the two possibilities together.
The | symbol requires the parenthesis because its precedence is very low, so you have to explicitly state the boundaries of the OR.
This regex relies heavily on zero-width assertions (the \A and \Z anchors, and the lookaround groups). You should always understand a regex before you use it for anything serious or permanent (otherwise you might catch a case of perl), so you might want to check out Start of String and End of String Anchors and Lookahead and Lookbehind Zero-Width Assertions.
Corrections welcome, of course!

If I understand the question correctly, you can use search/replace...just wildcard around your expression and then substitute the first and last parts.
s/^(.*)(your regex here)(.*)$/$1$3/

im not sure if this will work exactly as you intend but it might help:
Whatever you place in the brackets [] will be matched against. If you put ^ within the bracket, i.e [^a-zA-Z0-9/.] it will match everything except what is in the brackets.
http://www.regular-expressions.info/

Related

Having difficulty in a understanding regex backtracking

I was browsing through the regex tagged questions on SO when i came accross this problem,
A regex for a url was needed, the url begins with domain.com/advertorials/
The regex should match the following scenarios,
domain.com/advertorials
domain.com/advertorials?test=true
domain.com/advertorials/
domain.com/advertorials/?test=true
but not this,
domain.com/advertorials/version1?test=true
I came up with this regex advertorials\/?(?:(?!version)(.*))
This should work, but it doesnt for the last case. Looking at the debugger in regex101.com,
i see that after matching 's/' it matches 'version' word character by character and ultimately matches but since this is negative lookahead the condition fails. And this is the part i dont understand after failing it backtracks to before the '/' in 's/' and not after 's/'.
Is this how its supposed to work?? Can anyone help me understand?
(here's the demo link: https://regex101.com/r/ww3HR8/1).
Thanks,
Note: People already gave their solutions on that problem i just want to know why my regex fails.
The backtracking mechanism is in charge of this phenomenon, as you have already pointed out.
The ? quantifier, matching 1 or 0 repetitions of the quantified subpattern lets the regex engine match the string in two ways: either matching the quantified subpattern, or go on matching the string with subsequent subpattern.
So, advertorials/?(?!version)(.*) (I removed the redundant (?:...) non-capturing group), when applied to domain.com/advertorials/version1?test=true, matches advertorials, then matches /, and then the negative lookahead checks if, immediately to the right of the current position, there is version substring. Since there is version after /, the regex engine goes back and sees that /? pattern can match an empty string. So, the lookahead check is re-applied striaght after advertorials. There is no version after advertorials, and the match is returned.
The usual solution is using possessive quantifiers or atomic groups, but there are other approaches, too.
E.g.
advertorials\/?+(?!version)(.*)
^^
See the regex demo. Here, \/?+ matches 1 or 0 / chars, but once it matches, the egine cannot go back and re-match a part of a string with this pattern.
Or, you may include the /? in the lookahead and place it before /? pattern:
advertorials(?!\/?version)\/?(.*)
See another regex demo.
If you plan to disallow version anywhere after advertorials use
advertorials(?!.*version)\/?(.*)
See yet another demo.
Making the slash optional means there is a way to match without violating the constraint. If there is a way to match, the regex engine will find it, always.
Make the slash non-optional when it's followed by anything at all.
advertorials(?:/(?!version).*)?$
Incidentally, regex itself doesn't require the slash to be backslash-escaped (though some host languages use slashes as regex delimiters, so maybe you need to put it back). I also removed some redundant parentheses.
The reason:
This highlighted part is optional
advertorials\/?(?:(?!version)(.*))
Therefore it can also be advertorials(?:(?!version)(.*))
which matches advertorials/version
Essentially, (?!version)(.*) matches /version
Btw, this is normal backtracking by 1 character.
If you have already fixed it, then we're done !

Regex everything after, but not including

I am trying to regex the following string:
https://www.amazon.com/Tapps-Top-Apps-and-Games/dp/B00VU2BZRO/ref=sr_1_3?ie=UTF8&qid=1527813329&sr=8-3&keywords=poop
I want only B00VU2BZRO.
This substring is always going to be a 10 characters, alphanumeric, preceded by dp/.
So far I have the following regex:
[d][p][\/][0-9B][0-9A-Z]{9}
This matches dp/B00VU2BZRO
I want to match only B00VU2BZRO with no dp/
How do I regex this?
Here is one regex option which would produce an exact match of what you want:
(?<=dp\/)(.*)(?=\/)
Demo
Note that this solution makes no assumptions about the length of the path fragment occurring after dp/. If you want to match a certain number of characters, replace (.*) with (.{10}), for example.
Depending on your language/method of application, you have a couple of options.
Positive look behind. This will make your regex more complicated, but will make it match what you want exactly:
(<=dp/)[0-9A-Z]{10}
The construct (<=...) is called a positive look behind. It will not consume any of the string, but will only allow the match to happen if the pattern between the parens is matched.
Capture group. This will make the regex itself slightly simpler, but will add a step to the extraction process:
dp/([0-9A-Z]{10})
Anything between plain parens is a capture group. The entire pattern will be matched, including dp/, but most languages will give you a way of extracting the portion you are interested in.
Depending on your language, you may need to escape the forward slash (/).
As an aside, you never need to create a character class for single characters: [d][p][\/] can equally well be written as just dp\/.

regex look ahead behind (look around) negative problems

I am having trouble understanding negative regex lookahead / lookbehind. I got the impression from reading tutorials that when you set a criteria to look for, the criteria doesn't form part of the search match.
That seems to hold for positive lookahead examples I tried, but when I tried these negative ones, it matches the entire test string. 1, it shouldn't have matched anything, and 2 even if it did, it wasn't supposed to include the lookahead criteria??
(?<!^And).*\.txt$
with input
And.txt
See: https://regex101.com/r/vW0aXS/1
and
^A.*(?!\.txt$)
with input:
A.txt
See: https://regex101.com/r/70yeED/1
PS: if you're going to ask me which language. I don't know. we've been told to use regex without any specific reference to any specific languages. I tried clicking various options on regex101.com and they all came up the same.
Lookarounds only try to match at their current position.
You are using a lookbehind at the beginning of the string (?<!^And).*\.txt$, and a lookahead at the end of the string ^A.*(?!\.txt$), which won't work. (.* will always consume the whole string as it's first match)
To disallow "And", for example, you can put the lookahead at the beginning of the string with a greedy quantifier .* inside it, so that it scans the whole string:
(?!.*And).*\.txt$
https://regex101.com/r/1vF50O/1
Your understanding is correct and the issue is not with the lookbehind/lookahead. The issue is with .* which matches the entire string in both cases. The period . matches any character and then you follow it with * which makes it match the entire string of any length. Remove it and both you regexes will work:
(?<!^And)\.txt$
^A(?!\.txt$)

What mistake did I do for this unexpected negative lookahead subpattern?

I am actually working with a .tsv database whose headers are full of meaningful things for me.
I thus wanted to rip them off from the header to something that I & others users (non proficient with relational databases, so we mostly use Excel in the end to organize data and process it) would be more able to handle with Excel, by breaking them up with tabs.
Example header:
>(name1)database-ID:database2-ID:value1:value2
(I know this seems strange to put values in an header but this is descriptive of parameters of the third value associated to the header, that we don't have to mess here)
output as:
name1\tdatabase-ID\tdatabase2-ID\tvalue1\tvalue2\n
I thus pasted my data (headers, one per line) in EmEditor (BOOST syntax) and came with this regex:
>\((.*)\)(.*?)\:(.*?)\:(.*?)\:(.*?)\n
with each capturing group being then separated from others by inserting tabs between each others. It works, with perfect matches, no problem.
But I became aware there were malformed lines that didn't respected the logic of the whole database, and I wanted to make an expression to separate them at once.
If I make it with wrong lines it would be:
>(name1)database-ID:database2-ID:value1-1:value1-2\n
>(name2)database-ID:database2-ID:value2-1:value2-2\n
>(name3)database-ID:database2-ID:value3-1value3-2\n
Last line is ill-formed because it lacks the : between both last values.
I want it to be matched by working around the original expression that recognizes well-formed lines.
I perfectly know that I could came with different solutions by slightly tweaking my first expression for eliminating the good lines and retrieving misformed one after but
I don't want a solution to my process, I just want to understand what I made not well there; so that I become more educated (and not just more tricky by being able to circumvent my mistakes that I can't resolve):
I tried a negation of the above mentioned expression:
([^(>\((.*)\)(.*?)\:(.*?)\:(.*?)\:(.*?)\n)])
That doesn't match with anything.
I tried a negative lookahead, but It will be extremely, painfully slow then will match every 0-length matches possible in the document:
(?!(^>\((.*)\)(.*?)\:(.*?)\:(.*?)\:(.*?)\n))
I thus added a group capture for a string of characters behind,
but it doesn't work either:
(?!(^>\((.*)\)(.*?)\:(.*?)\:(.*?)\:(.*?)\n))(^.*?)
So please explain me where I have been wrong with the negating group ([^whatever]) and the use of the negative lookahead?
So please explain me where I have been wrong with the negating group ([^whatever]) and the use of the negative lookahead?
Let's address the question first: What does [^(pattern)] do?
You seem to have a misunderstanding and expect it to:
Match everything except the subpattern pattern. (Negation)
What it actually does is to:
Match any character that aren't (, p, a, t, ... n, ).
Therefore, the pattern
([^(>\((.*)\)(.*?)\:(.*?)\:(.*?)\:(.*?)\n)])
... Matches a character that aren't (, >, (, ... \n, ).
As for the negative lookahead, you're simply doing it wrong. The anchor ^ is in the wrong position, therefore your assertion will fail to provide any useful help. It's also not what negative lookaheads are for altogether.
(?!(^>\((.*)\)(.*?)\:(.*?)\:(.*?)\:(.*?)\n))
I'll explain what this does:
(?! Open negative lookahead group: Assert the position does not match this pattern, without moving the pointer position.
( Capturing group. The use of capturing groups in negative lookaheads are useless, as the subpattern in negative lookahead groups never matches.
^ Assert position at start of string.
>\( Literal character sequence ">(".
(.*) Capturing group which matches as many characters as possible except newlines, then backtracks.
\) Literal character ")".
(.*?) Capturing group with reluctant zero-to-one match of any characters except newlines.
\: Literal character ":".
(.*?)\:(.*?)\:(.*?)
\n A new line.
) Closes capturing group.
) Closes negative lookahead group. When this assertion is finished, the pointer position is same as beginning, and thus the resulting match is zero-length.
Note that the anchor is nested within the negative lookahead group. It should be at the start:
^(?!(>\((.*)\)(.*?)\:(.*?)\:(.*?)\:(.*?)\n))
While this doesn't return anything useful, it explains what is wrong, since you don't need a solution. ;)
In case you are in need of a solution suddenly, please refer to this relevant answer of mine (I'm not adding anything else into the post):
Rails 3 - Precompiling all css, sass and scss files in a folder
You could do this simply through PCRE Verb (*SKIP)(*F). The below regex would match all the bad-lines.
(?:^>\([^()]*\):[^:]*:[^:]*:[^:]*:[^:\n]*$)(*SKIP)(*F)|^.+
DEMO
Based on what I have been reading from Unihedron;
This is what I came for in emEditor:
^(?!>\(([A-Za-z0-9_\'\-]*?)\)(([A-Za-z0-9_\'\-]*?)\:){3}([A-Za-z0-9_\'\-]*?)\n).*\n
>(name1)database-ID:database2-ID:value1-1:value1-2
(NOT MATCH)
>(name2)database-ID:database2-ID:value2-1:value2-2
(NOT MATCH)
>(name3)database-ID:database2-ID:value3-1value3-2
(MATCH)
>(name3)database-ID::database2-ID:value3-1:value3-2
(MATCH)
(the character class avoid discarding names including special characters without making it possible to have two subsequent ":".)
I also could achieve the same results with:
(?!^>\(([A-Za-z0-9_\'\-]*?)\)(([A-Za-z0-9_\'\-]*?)\:){3}([A-Za-z0-9_\'\-]*?)\n)^.*\n
So I guess that all along capturing groups were what was messing with my lookahead.
Now I acknowledge that Avinash Raj is more efficient with the (*SKIP)(*F)|^.+ pattern, just that I didn't know about those functions and I also wanted to understand my logic / syntax mistake. (Thanks to Unihedron for that)

Google Analytics Regex - Alternative to no negative lookahead

Google Analytics does not allow negative lookahead anymore within its filters. This is proving to be very difficult to create a custom report only including the links I would like it to include.
The regex that includes negative lookahead that would work if it was enabled is:
test.com(\/\??index\_(.*)\.php\??(.*)|\/\?(.*)|\/|)+(\s)*(?!.)
This matches:
test.com
test.com/
test.com/index_fb2.php
test.com/index_fb2.php?ref=23
test.com/index_fb2.php?ref=23&e=35
test.com/?ref=23
test.com/?ref=23&e=35
and does not match (as it should):
test.com/ambassadors
test.com/admin/?signup=true
test.com/randomtext/
I am looking to find out how to adapt my regex to still hold the same matches but without the use of negative lookahead.
Thank you!
Google Analytics doesn't seem to support single-line and multiline modes, which makes sense to me. URLs can't contain newlines, so it doesn't matter if the dot doesn't match them and there's never any need for ^ and $ to match anywhere but the beginning and end of the whole string.
That means the (?!.) in your regex is exactly equivalent to $, which matches only at the very end of the string (like \z, in flavors that support it). Since that's the only lookahead in your regex, you should never have have had this problem; you should have been using $ all along.
However, your regex has other problems, mostly owing to over-reliance on (.*). For example, it matches these strings:
test.com/?^#(%)!*%supercalifragilisticexpialidocious
test.com/index_ecky-ecky-ecky-ecky-PTANG!-vroop-boing_rowr.php (ni! shh!)
...which I'm pretty sure you don't want. :P
Try this regex:
test\.com(?:/(?:index_\w+\.php)?(?:\?ref=\d+(?:&e=\d+)?)?)?\s*$
or more readably:
test\.com
(?:
/
(?:index_\w+\.php)?
(?:
\?ref=\d+
(?:
&e=\d+
)?
)?
)?
\s*$
For illustration purposes I'm making a lot of simplifying assumptions about (e.g.) what parameters can be present, what order they'll appear in, and what their values can be. I'm also wondering if it's really necessary to match the domain (test.com). I have no experience with Google Analytics, but shouldn't the match start (and be anchored) right after domain? And do you really have to allow for whitespace at the end? It seems to me the regex should be more like this:
^/(?:index_\w+\.php)?(?:\?ref=\d+(?:&e=\d+)?)?$
Firstly I think your regex needs some fixing. Let's look at what you have:
test.com(\/\??index_.*.php\??(.*)|\/\?(.*)|\/|)+(\s)*(?!.)
The case where you use the optional ? at the start of index... is already taken care of by the second alternative:
test.com(\/index_.*.php\??(.*)|\/\?(.*)|\/|)+(\s)*(?!.)
Now you probably only want the first (.*) to be allowed, if there actually was a literal ? before. Otherwise you will match test.com/index_fb2.phpanystringhereandyouprobablydon'twantthat. So move the corresponding optional marker:
test.com(\/index_.*.php(\?(.*))?|\/\?(.*)|\/|)+(\s)*(?!.)
Now .* consumes any character and as much as possible. Also, the . in front of php consumes any character. This means you would be allowing both test.com/index_fb2php and test.com/index_fb2.html?someparam=php. Let's make that a literal . and only allow non-question-mark characters:
test.com(\/index_[^?]*\.php(\?(.*))?|\/\?(.*)|\/|)+(\s)*(?!.)
Now the first and second and third option can be collapsed into one, if we make the file name optional, too:
test.com(\/(index_[^?]*\.php)?(\?(.*))?|)+(\s)*(?!.)
Finally, the + can be removed, because the (.*) inside can already take care of all possible repetitions. Also (something|) is the same as (something)?:
test.com(\/(index_[^?]*\.php)?(\?(.*))?)?(\s)*(?!.)
Seeing your input examples, this seems to be closer to what you actually want to match.
Then to answer your question. What (?!.) does depends on whether you use singleline mode or not. If you do, it asserts that you have reached the end of the string. In this case you can simply replace it by \Z, which always matches the end of the string. If you do not, then it asserts that you have reached the end of a line. In this case you can use $ but you need to also use multi-line mode, so that $ matches line-endings, too.
So, if you use singleline mode (which probably means you have only one URL per string), use this:
test.com(\/(index_[^?]*\.php)?(\?(.*))?)?(\s)*\Z
If you do not use singleline mode (which probably means you can have multiple URLs on their own lines), you should also use multiline mode and this kind of anchor instead:
test.com(\/(index_[^?]*\.php)?(\?(.*))?)?(\s)*$