CODEWITHSUNDEEP
c#csvparsingmalformedtextfieldparser
eddie_cat
eddie_cat

Reputation: 2583

How can I deal with parsing bad csv data?

I know that the data should be correct. I have no control over the data and my boss is just going to tell me that I need to figure out a way to deal with someone else's mistake. So please don't tell me it's not my problem that the data is bad, because it is.

Anywho, this is what I'm looking at:

"Words","[email protected]","","4253","57574","FirstName","","LastName, MD","","","576JFJD","","1971","","Words","Address","SUITE "A"","City","State","Zip","Phone","",""

Data has been scrubbed for confidentiality reasons.

So as you see, the data contains quotation marks and there are commas inside some of these quoted fields. So I cannot remove them. But the "Suite A""" is throwing off the parser. There are too many quotation marks. >.<

I'm using the TextFieldParser in the Microsoft.VisualBasic.FileIO namespace with these settings:

            parser.HasFieldsEnclosedInQuotes = true;
            parser.SetDelimiters(",");
            parser.TextFieldType = FieldType.Delimited;

The error is

MalformedLineException: Line 9871 cannot be parsed using the current delimiters.

I would like to scrub the data somehow to account for this but I'm not sure how to do it. Or maybe there's a way to just skip this line? Although I suspect my higher ups will not approve of me just skipping data that we might need.

Upvotes: 0

Views: 4386

Answers (6)

TheHans255
TheHans255

Reputation: 2223

At the core of CSV's file format, each line is a row, each cell in that row is separated by a comma. In your case, your format also contains the (very unfortunate) stipulation that commas inside a pair of quotation marks do not count as separators and are instead part of the data. I say very unfortunate because a misplaced quotation mark affects the entire rest of the line, and since quotation marks in standard ASCII do not distinguish between open and closed, there really is nothing you can do to recover from this without knowing the original intent.

That is when you log a message in a way that the person who does know the original intent (the person that provided the data) can look at the file and correct the error:

if (parse_line(line, &data)) {
   // save the data
} else {
   // log the error
   fprintf(&stderr, "Bad line: %s", line);
}

And since your quotation marks aren't escaping newlines, you can keep on going with the next line after running into this error.

ADDENDUM: And if your company has a choice (i.e. your data is being serialized by a company tool) don't use CSV. Use something like XML or JSON with a much more clearly defined parsing mechanism.

Upvotes: 1

Shannon Holsinger
Shannon Holsinger

Reputation: 2341

My only addition to what everyone is saying (because we've all been there) is to try to attempt to rectify each new issue you encounter with code. There are some decent REGEX strings out there https://www.google.com/?ion=1&espv=2#q=c-sharp+regex+csv+clean or you could manually fix things using String.Replace (String.Replace("\"\"\"","").Replace("\"\","").Replace("\",,","\",") or such). Eventually, as you detect and find ways of correcting more and more mistakes, your manual recovery rate will be minimized substantially (most of your bad data will likely come from similar mistakes). Cheers!

PS - Idea-ish (it's been a while - the logic may neeed some tweaking as I'm writing from memory), but you'll get the gist:

public string[] parseCSVWithQuotes(string csvLine,int expectedNumberOfDataPoints)
    {
        string ret = "";
        string thisChar = "";
        string lastChar = "";
        bool needleDown = true;
        for(int i = 0; i < csvLine.Length; i++)
        {
            thisChar = csvLine.Substring(i, 1);
            if (thisChar == "'"&&lastChar!="'")
                needleDown = needleDown == true ? false : true;//when needleDown = true, characters are treated literally
            if (thisChar == ","&&lastChar!=",") {
                if (needleDown)
                {
                    ret += "|";//convert literal comma to pipe so it doesn't cause another break on split
                }else
                {
                    ret += ",";//break on split is intended because the comma is outside the single quote
                }
            }
            if (!needleDown && (thisChar == "\"" || thisChar == "*")) {//repeat for any undesired character or use RegEx
                                                                       //do not add -- this eliminates any undesired characters outside single quotes
            }
            else
            {
                if ((lastChar == "'" || lastChar == "\"" || lastChar == ",") && thisChar == lastChar)
                {
                    //do not add - this eliminates double characters
                }else
                {
                    ret += thisChar;
                    lastChar = thisChar;
                    //this character is not an undesired character, is no a double, is valid.
                }
            }
        }
        //we've cleaned as best we can
        string[] parts = ret.Split(',');
        if(parts.Length==expectedNumberOfDataPoints){
        for(int i = 0; i < parts.Length; i++)
        {
            //go back and replace the temporary pipe with the literal comma AFTER split
            parts[i] = parts[i].Replace("|", ",");
        }

        return parts;
        }else{
            //save ret to bad CSV log
            return null;
        }
    }

Upvotes: 2

PapaAtHome
PapaAtHome

Reputation: 604

I had to do this once aswell. My approach was to go through a line and keep track on what I was reading. Basicly, I coded my own scanner chopping off tokens from the input line which gave me full control over my faulty .csv data.

This is what I did:

For each character on a line of input.
 1. when outside of a string meeting a comma => all of the previous string (which can be empty) is a valid token.
 2. when outside of a sting meeting anything but a comma or a quote => now you have a real problem, unquoted tekst => handle as you see fit.
 3. when outside of a string meeing a quote => found a start of string.
 4. when inside of a string meeting a comma => accept the comma as part of the string.
 5. when inside of the string meeting a qoute => trouble starts here, mark this point.
   6. continue and when meeting a comma (skipping white space if desired) close the string, 'unread' the comma and continue. (than will bring you to point 1.)
   7. or continue and when meeting a quote -> obviously, what was read must be part of the string, add it to the string, 'unread' the quote and continue. (that will you bring to point 5)
   8. or continue and find an whitespace, then End Of Line ('\n') -> the last qoute must be the closing quote. accept the string as a value.
   9. or continue and fine non-whitespace, then End Of Line. -> now you have a real problem, you have the start of a string but it is not closed -> handle the error as you see fit.

If the number of fields in your .csv file is fixed you can count the comma's you recognise as field seperators and when you see a End Of Line you know you have another problem or not.

With the stream of strings received from the input line you can build a 'clean' .csv line and this way build a buffer of accepted and cleaned input that you can use in your already existing code.

Upvotes: 0

Pat
Pat

Reputation: 2750

If you are only trying to get rid of the stray " marks in your csv, you can use the following regex to find them and replace them with ' 

String sourcestring = "source string to match with pattern";
String matchpattern = @"(?<!^|,)""(?!(,|$))";
String replacementpattern = @"$1'";
Console.WriteLine(Regex.Replace(sourcestring,matchpattern,replacementpattern,RegexOptions.Multiline));

Explanation:

@"(?<!^|,)""(?!(,|$))"; will find will find any " that is not preceded by the beginning of the string, or a , and that is not followed by the end of the string or a ,

Upvotes: 3

Sam I am says Reinstate Monica
Sam I am says Reinstate Monica

Reputation: 31194

I've had to do this before,

The first step is to parse the data using string.split(',')

The next step is to combine the segments that belong together.

What I essentially did was

  • make a new list representing the combined strings
  • if a string begins with a quote, push it onto your new list
  • if it does not begin with a quote, append it to the last string in your list
  • Bonus: throw exceptions when a string ends with a quote but the next one does not begin with a quote

Depending on what the rules are regarding what can actually appear in your data, you might have to change your code to account for that.

Upvotes: 1

Xiaoy312
Xiaoy312

Reputation: 14477

I am not familiar with TextFieldParser. However with CsvHelper, you can add a custom handler for invalid data:

var config = new CsvConfiguration();
config.IgnoreReadingExceptions = true;
config.ReadingExceptionCallback += (e, row) =>
{
    // you can add some custom patching here if possible
    // or, save the line numbers and add/edit them manually later.
};

using(var file = File.OpenRead(".csv"))
using(var reader = new CsvReader(reader, config))
{
    reader.GetRecords<YourDtoClass>();
}

Upvotes: 2

Related Questions