I have a file with many lines including a string like this: blah blah num=12345; blah blah
I would like to find lines where the number after the equals sign is greater than 1, with no upper limit. (I do not expect a number to ever start with zero.)
I started with this expression that will match any number starting with any digit that's not a 1, and it works fine and I understand it.
grep 'num=[2-9][0-9]*;'
This next expression should, I thought, return any number starting with a 1 that has two or more digits, but I instead get nothing back:
grep 'num=1[0-9]+;'
I though the above meant: must match num=1, then must match something between 0-9 one or more times, then must match ;. Where am I going wrong?
With grep you must escape the + quantifier
grep 'num=1[0-9]\+;'
For your problem you can use this (for all numbers >1, if i understand well):
grep 'num=\([2-9]\|1[0-9]\)[0-9]*;'
Is it possible to increment numbers using regex substitution? Not using evaluated/function-based substitution, of course.
This question was inspired by another one, where the asker wanted to increment numbers in a text editor. There are probably more text editors that support regex substitution than ones that support full-on scripting, so a regex might be convenient to float around, if one exists.
Also, often I've learned neat things from clever solutions to practically useless problems, so I'm curious.
Assume we're only talking about non-negative decimal integers, i.e. \d+.
Is it possible in a single substitution? Or, a finite number of substitutions?
If not, is it at least possible given an upper bound, e.g. numbers up to 9999?
Of course it's doable given a while-loop (substituting while matched), but we're going for a loopless solution here.
This question's topic amused me for one particular implementation I did earlier. My solution happens to be two substitutions so I'll post it.
My implementation environment is solaris, full example:
echo "0 1 2 3 7 8 9 10 19 99 109 199 909 999 1099 1909" |
perl -pe 's/\b([0-9]+)\b/0$1~01234567890/g' |
perl -pe 's/\b0(?!9*~)|([0-9])(?=9*~[0-9]*?\1([0-9]))|~[0-9]*/$2/g'
1 2 3 4 8 9 10 11 20 100 110 200 910 1000 1100 1910
Pulling it apart for explanation:
s/\b([0-9]+)\b/0$1~01234567890/g
For each number (#) replace it with 0#~01234567890. The first 0 is in case rounding 9 to 10 is needed. The 01234567890 block is for incrementing. The example text for "9 10" is:
09~01234567890 010~01234567890
The individual pieces of the next regex can be described seperately, they are joined via pipes to reduce substitution count:
s/\b0(?!9*~)/$2/g
Select the "0" digit in front of all numbers that do not need rounding and discard it.
s/([0-9])(?=9*~[0-9]*?\1([0-9]))/$2/g
(?=) is positive lookahead, \1 is match group #1. So this means match all digits that are followed by 9s until the '~' mark then go to the lookup table and find the digit following this number. Replace with the next digit in the lookup table. Thus "09~" becomes "19~" then "10~" as the regex engine parses the number.
s/~[0-9]*/$2/g
This regex deletes the ~ lookup table.
Wow, turns out it is possible (albeit ugly)!
In case you do not have the time or cannot be bothered to read through the whole explanation, here is the code that does it:
$str = '0 1 2 3 4 5 6 7 8 9 10 11 12 13 19 20 29 99 100 139';
$str = preg_replace("/\d+/", "$0~", $str);
$str = preg_replace("/$/", "#123456789~0", $str);
do
{
$str = preg_replace(
"/(?|0~(.*#.*(1))|1~(.*#.*(2))|2~(.*#.*(3))|3~(.*#.*(4))|4~(.*#.*(5))|5~(.*#.*(6))|6~(.*#.*(7))|7~(.*#.*(8))|8~(.*#.*(9))|9~(.*#.*(~0))|~(.*#.*(1)))/s",
"$2$1",
$str, -1, $count);
} while($count);
$str = preg_replace("/#123456789~0$/", "", $str);
echo $str;
Now let's get started.
So first of all, as the others mentioned, it is not possible in a single replacement, even if you loop it (because how would you insert the corresponding increment to a single digit). But if you prepare the string first, there is a single replacement that can be looped. Here is my demo implementation using PHP.
I used this test string:
$str = '0 1 2 3 4 5 6 7 8 9 10 11 12 13 19 20 29 99 100 139';
First of all, let's mark all digits we want to increment by appending a marker character (I use ~, but you should probably use some crazy Unicode character or ASCII character sequence that definitely will not occur in your target string.
$str = preg_replace("/\d+/", "$0~", $str);
Since we will be replacing one digit per number at a time (from right to left), we will just add that marking character after every full number.
Now here comes the main hack. We add a little 'lookup' to the end of our string (also delimited with a unique character that does not occur in your string; for simplicity I used #).
$str = preg_replace("/$/", "#123456789~0", $str);
We will use this to replace digits by their corresponding successors.
Now comes the loop:
do
{
$str = preg_replace(
"/(?|0~(.*#.*(1))|1~(.*#.*(2))|2~(.*#.*(3))|3~(.*#.*(4))|4~(.*#.*(5))|5~(.*#.*(6))|6~(.*#.*(7))|7~(.*#.*(8))|8~(.*#.*(9))|9~(.*#.*(~0))|(?<!\d)~(.*#.*(1)))/s",
"$2$1",
$str, -1, $count);
} while($count);
Okay, what is going on? The matching pattern has one alternative for every possible digit. This maps digits to successors. Take the first alternative for example:
0~(.*#.*(1))
This will match any 0 followed by our increment marker ~, then it matches everything up to our cheat-delimiter and the corresponding successor (that is why we put every digit there). If you glance at the replacement, this will get replaced by $2$1 (which will then be 1 and then everything we matched after the ~ to put it back in place). Note that we drop the ~ in the process. Incrementing a digit from 0 to 1 is enough. The number was successfully incremented, there is no carry-over.
The next 8 alternatives are exactly the same for the digits 1to 8. Then we take care of two special cases.
9~(.*#.*(~0))
When we replace the 9, we do not drop the increment marker, but place it to the left of our the resulting 0 instead. This (combined with the surrounding loop) is enough to implement carry-over propagation. Now there is one special case left. For all numbers consisting solely of 9s we will end up with the ~ in front of the number. That is what the last alternative is for:
(?<!\d)~(.*#.*(1))
If we encounter a ~ that is not preceded by a digit (therefore the negative lookbehind), it must have been carried all the way through a number, and thus we simply replace it with a 1. I think we do not even need the negative lookbehind (because this is the last alternative that is checked), but it feels safer this way.
A short note on the (?|...) around the whole pattern. This makes sure that we always find the two matches of an alternative in the same references $1 and $2 (instead of ever larger numbers down the string).
Lastly, we add the DOTALL modifier (s), to make this work with strings that contain line breaks (otherwise, only numbers in the last line will be incremented).
That makes for a fairly simple replacement string. We simply first write $2 (in which we captured the successor, and possibly the carry-over marker), and then we put everything else we matched back in place with $1.
That's it! We just need to remove our hack from the end of the string, and we're done:
$str = preg_replace("/#123456789~0$/", "", $str);
echo $str;
> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 20 21 30 100 101 140
So we can do this entirely in regular expressions. And the only loop we have always uses the same regex. I believe this is as close as we can get without using preg_replace_callback().
Of course, this will do horrible things if we have numbers with decimal points in our string. But that could probably be taken care of by the very first preparation-replacement.
Update: I just realised, that this approach immediately extends to arbitrary increments (not just +1). Simply change the first replacement. The number of ~ you append equals the increment you apply to all numbers. So
$str = preg_replace("/\d+/", "$0~~~", $str);
would increment every integer in the string by 3.
I managed to get it working in 3 substitutions (no loops).
tl;dr
s/$/ ~0123456789/
s/(?=\d)(?:([0-8])(?=.*\1(\d)\d*$)|(?=.*(1)))(?:(9+)(?=.*(~))|)(?!\d)/$2$3$4$5/g
s/9(?=9*~)(?=.*(0))|~| ~0123456789$/$1/g
Explanation
Let ~ be a special character not expected to appear anywhere in the text.
If a character is nowhere to be found in the text, then there's no way to make it appear magically. So first we insert the characters we care about at the very end.
s/$/ ~0123456789/
For example,
0 1 2 3 7 8 9 10 19 99 109 199 909 999 1099 1909
becomes:
0 1 2 3 7 8 9 10 19 99 109 199 909 999 1099 1909 ~0123456789
Next, for each number, we (1) increment the last non-9 (or prepend a 1 if all are 9s), and (2) "mark" each trailing group of 9s.
s/(?=\d)(?:([0-8])(?=.*\1(\d)\d*$)|(?=.*(1)))(?:(9+)(?=.*(~))|)(?!\d)/$2$3$4$5/g
For example, our example becomes:
1 2 3 4 8 9 19~ 11 29~ 199~ 119~ 299~ 919~ 1999~ 1199~ 1919~ ~0123456789
Finally, we (1) replace each "marked" group of 9s with 0s, (2) remove the ~s, and (3) remove the character set at the end.
s/9(?=9*~)(?=.*(0))|~| ~0123456789$/$1/g
For example, our example becomes:
1 2 3 4 8 9 10 11 20 100 110 200 910 1000 1100 1910
PHP Example
$str = '0 1 2 3 7 8 9 10 19 99 109 199 909 999 1099 1909';
echo $str . '<br/>';
$str = preg_replace('/$/', ' ~0123456789', $str);
echo $str . '<br/>';
$str = preg_replace('/(?=\d)(?:([0-8])(?=.*\1(\d)\d*$)|(?=.*(1)))(?:(9+)(?=.*(~))|)(?!\d)/', '$2$3$4$5', $str);
echo $str . '<br/>';
$str = preg_replace('/9(?=9*~)(?=.*(0))|~| ~0123456789$/', '$1', $str);
echo $str . '<br/>';
Output:
0 1 2 3 7 8 9 10 19 99 109 199 909 999 1099 1909
0 1 2 3 7 8 9 10 19 99 109 199 909 999 1099 1909 ~0123456789
1 2 3 4 8 9 19~ 11 29~ 199~ 119~ 299~ 919~ 1999~ 1199~ 1919~ ~0123456789
1 2 3 4 8 9 10 11 20 100 110 200 910 1000 1100 1910
Is it possible in a single substitution?
No.
If not, is it at least possible in a single substitution given an upper bound, e.g. numbers up to 9999?
No.
You can't even replace the numbers between 0 and 8 with their respective successor. Once you have matched, and grouped this number:
/([0-8])/
you need to replace it. However, regex doesn't operate on numbers, but on strings. So you can replace the "number" (or better: digit) with twice this digit, but the regex engine does not know it is duplicating a string that holds a numerical value.
Even if you'd do something (silly) as this:
/(0)|(1)|(2)|(3)|(4)|(5)|(6)|(7)|(8)/
so that the regex engine "knows" that if group 1 is matched, the digit '0' is matched, it still cannot do a replacement. You can't instruct the regex engine to replace group 1 with the digit '1', group '2' with the digit '2', etc. Sure, some tools like PHP will let you define a couple of different patterns with corresponding replacement strings, but I get the impression that is not what you were thinking about.
It is not possible by regular expression search and substitution alone.
You have to use use something else to help achieve that. You have to use the programming language at hand to increment the number.
Edit:
The regular expressions definition, as part of Single Unix Specification doesn't mention regular expressions supporting evaluation of aritmethic expressions or capabilities for performing aritmethic operations.
Nonetheless, I know some flavors ( TextPad, editor for Windows) allows you to use \i as a substitution term which is an incremental counter of how many times has the search string been found, but it doesn't evaluate or parse found strings into a number nor does it allow to add a number to it.
I have found a solution in two steps (Javascript) but it relies on indefinite lookaheads, which some regex engines reject:
const incrementAll = s =>
s.replaceAll(/(.+)/gm, "$1\n101234567890")
.replaceAll(/(?:([0-8]|(?<=\d)9)(?=9*[^\d])(?=.*\n\d*\1(\d)\d*$))|(?<!\d)9(?=9*[^\d])(?=(?:.|\n)*(10))|\n101234567890$/gm, "$2$3");
The key thing is to add a list of numbers in order at the end of the string in the first step, and in the second, to find the location relevant digit and capture the digit to its right via a lookahead. There are two other branches in the second step, one for dealing with initial nines, and the other for removing the number sequence.
Edit: I just tested it in safari and it throws an error, but it definately works in firefox.
I needed to increment indices of output files by one from a pipeline I can't modify. After some searches I got a hit on this page. While the readings are meaningful, they really don't give a readable solution to the problem. Yes it is possible to do it with only regex; no it is not as comprehensible.
Here I would like to give a readable solution using Python, so that others don't need to reinvent the wheels. I can imagine many of you may have ended up with a similar solution.
The idea is to partition file name into three groups, and format your match string so that the incremented index is the middle group. Then it is possible to only increment the middle group, after which we piece the three groups together again.
import re
import sys
import argparse
from os import listdir
from os.path import isfile, join
def main():
parser = argparse.ArgumentParser(description='index shift of input')
parser.add_argument('-r', '--regex', type=str,
help='regex match string for the index to be shift')
parser.add_argument('-i', '--indir', type=str,
help='input directory')
parser.add_argument('-o', '--outdir', type=str,
help='output directory')
args = parser.parse_args()
# parse input regex string
regex_str = args.regex
regex = re.compile(regex_str)
# target directories
indir = args.indir
outdir = args.outdir
try:
for input_fname in listdir(indir):
input_fpath = join(indir, input_fname)
if not isfile(input_fpath): # not a file
continue
matched = regex.match(input_fname)
if matched is None: # not our target file
continue
# middle group is the index and we increment it
index = int(matched.group(2)) + 1
# reconstruct output
output_fname = '{prev}{index}{after}'.format(**{
'prev' : matched.group(1),
'index' : str(index),
'after' : matched.group(3)
})
output_fpath = join(outdir, output_fname)
# write the command required to stdout
print('mv {i} {o}'.format(i=input_fpath, o=output_fpath))
except BrokenPipeError:
pass
if __name__ == '__main__': main()
I have this script named index_shift.py. To give an example of the usage, my files are named k0_run0.csv, for bootstrap runs of machine learning models using parameter k. The parameter k starts from zero, and the desired index map starts at one. First we prepare input and output directories to avoid overriding files
$ ls -1 test_in/ | head -n 5
k0_run0.csv
k0_run10.csv
k0_run11.csv
k0_run12.csv
k0_run13.csv
$ ls -1 test_out/
To see how the script works, just print its output:
$ python3 -u index_shift.py -r '(^k)(\d+?)(_run.+)' -i test_in -o test_out | head -n5
mv test_in/k6_run26.csv test_out/k7_run26.csv
mv test_in/k25_run11.csv test_out/k26_run11.csv
mv test_in/k7_run14.csv test_out/k8_run14.csv
mv test_in/k4_run25.csv test_out/k5_run25.csv
mv test_in/k1_run28.csv test_out/k2_run28.csv
It generates bash mv command to rename the files. Now we pipe the lines directly into bash.
$ python3 -u index_shift.py -r '(^k)(\d+?)(_run.+)' -i test_in -o test_out | bash
Checking the output, we have successfully shifted the index by one.
$ ls test_out/k0_run0.csv
ls: cannot access 'test_out/k0_run0.csv': No such file or directory
$ ls test_out/k1_run0.csv
test_out/k1_run0.csv
You can also use cp instead of mv. My files are kinda big, so I wanted to avoid duplicating them. You can also refactor how many you shift as input argument. I didn't bother, cause shift by one is most of my use cases.
This command
$ echo '一二三四五六七八九十' | grep -oE '[一-十]'
outputs:
一
二
三
五
六
七
八
九
十
The regex [一-十] (one to ten) is expected to match against Chinese numbers.
As the example shows, it matches against every Chinese number from one to ten, except the Chinese character 四 (four)。
Why?
Is this a bug or a joke?
I may think this as joke, because in Chinese '四' (four) sounds alike '事' (thing). In fact, in some dialects of Chinese, they share a same pronunciation. Thus '一二三五六七八九十' (one two three five six seven eight nine ten) implies '沒四' (no four), i.e. '沒事' (no thing)。
BTW, the version of the grep I use:
GNU grep 2.5.4
The Chinese numbers are not in order in Unicode. That 四 is U+56DB, while 一 is U+4E00, and 10 is 5341. So the 4 doesn't fit.
Read the Unicode standard for more information, and see http://www.unicode.org/charts/PDF/U4E00.pdf.
I asked this on a Chinese forum, and alvin_rxg replies with this:
一 => U+4E00
十 => U+5341
四 => U+56DB
So I known I just made a silly assumption!
grep is not that smart.
The regex [一-十] (one to ten) is not expected to match against Chinese numbers.
It just matches any characters between '一' (U+4E00) and '十' (U+5341).
It just happens that all Chinese numbers are in that range except '四' (U+56DB)!
This can be used as a cold joke, though. ;-)
I want to find all numbers that are in between 500-599. I'm very new to regex, I came up with this :
5[0-9][0-9]+
This is working fine, matching 566,577,500. But it also matches 6578. Which I don't want.
Edit:
Here is my file contents:
asd 554
sad
sads
dsa
456
sa
d
dsa
asda
d500
521
519 asdasd
524 asdasdsdsadsdasd sadsadsadasdsd asdsa dsa dsadsad sad asdas dsa sad sad asds a 543
As many suggested I tried :
grep "^5[0-9]{2}$" test
which isn't finding any numbers at all!
How do I put a constraint on this?
If you want to match 5xx only on a line, and not when 5xx occurs as a part of x5xx,
^5\d{2}$
\d = Digit
^ = beginning of line
$ = end of line
EDIT:
Based on additional details in the question, you have a variable number of spaces at the beginning of the line, so, you want the following instead:
\s*5\d{2}\s
Matches spaces on either side of 5xx.
With grep the easiest way is to use -w to only match whole words:
grep --color=always -w "5[0-9][0-9]" test
Remove the + sign:
5[0-9][0-9]
This will match "5" succeeded by two numbers, and nothing else.
You have to describe a bit more accurately what you want to happen with e.g. 6578? If you want 578 in the output (because after "6" there is a sequence of characters matching your format 5xx) you can simply do
grep -o "5[0-9][0-9]"
Note that unlike other answers, the -o flag emits multiple numbers from a single line if needed.
If, on the other hand, you want to match words of format 5xx, you can add -w flag, too:
grep -o -w "5[0-9][0-9]"
For more complex rules for matching, you want to use -E flag instead and use possibly a much more complex regex.
Here's a sample string which I want do a regex on
101-nocola_conte_-_fuoco_fatuo_(koop_remix)
The first digit in "101" is the disc number and the next 2 digits are the track numbers. How do I match the track numbers and ignore the disc number (first digit)?
Something like
/^\d(\d\d)/
Would match one digit at the start of the string, then capture the following two digits
Do you mean that you don't mind what the disk number is, but you want to match, say, track number 01 ?
In perl you would match it like so: "^[0-9]01.*"
or more simply "^.01.*" - which means that you don't even mind if the first char is not a digit.
^\d(\d\d)
You may need \ in front of the ( depending on which environment you intend to run the regex into (like vi(1)).
Which programming language? For the shell something with egrep will do the job:
echo '101-nocola_conte_-_fuoco_fatuo_(koop_remix)' | egrep -o '^[0-9]{3}' | egrep -o '[0-9]{2}$'