Auto Detect CSV Separator




CSV

Introduction

CSV files are very popular for storing tabular data because they are simple textual files with very few rules. This makes them very interoperable because CSV readers and writers are relatively easy to implement. Interoperability is, probably, the first reason why someone would choose to save the data in a CSV format.

Although rules for writing and reading CSV files, which are explained in the next chapter, are relatively known and widely accepted, one rule is an exception; determining the character used as a separator. CSV files, as the name Comma Separated Values says, should use a comma (",") as the separator but there are many CSV files that use a semicolon (";") or a horizontal tab ("\t") as a separator.

So, in order to build a generic CSV reader that will read CSV files regardless of the separator, a reader must first figure out which character is used as a separator. This article gives one possible solution to this problem.

CSV format

Rules for writing CSV files are pretty simple:

  • If a value contains a separator character or a new line character or begins with a quote - enclose the value in quotes,
  • If a value is enclosed in quotes - any quote character contained in the value should be followed by an additional quote character.

These two simple rules enable us to write CSV writers easily, in just a few minutes. Implementing a CSV reader is much more problematic because a CSV stream has to be parsed sequentially, character by character and additional state storage has to be provided - which effectively makes a CSV reader a state machine. There are many CSV readers out there that have the wrong implementation because they do not follow the rules stated above.

Implementation

Now that we have defined the rules for CSV files, we can implement a CSV reader that is able to determine which character is used as a separator.

Here is an entire C# source code of the method that detects the separator in a CSV stream:

public static char Detect(TextReader reader, int rowCount, IList<char> separators)
{
    IList<int> separatorsCount = new int[separators.Count];

    int character;

    int row = 0;

    bool quoted = false;
    bool firstChar = true;

    while (row < rowCount)
    {
        character = reader.Read();

        switch (character)
        {
            case '"':
                if (quoted)
                {
                    if (reader.Peek() != '"') // Value is quoted and current character is " and next character is not ".
                        quoted = false;
                    else
                        reader.Read(); // Value is quoted and current and next characters are "" - read (skip) peeked qoute.
                }
                else
                {
                    if (firstChar) // Set value as quoted only if this quote is the first char in the value.
                        quoted = true;
                }
                break;
            case '\n':
                if (!quoted)
                {
                    ++row;
                    firstChar = true;
                    continue;
                }
                break;
            case -1:
                row = rowCount;
                break;
            default:
                if (!quoted)
                {
                    int index = separators.IndexOf((char)character);
                    if (index != -1)
                    {
                        ++separatorsCount[index];
                        firstChar = true;
                        continue;
                    }
                }
                break;
        }
 
        if (firstChar)
            firstChar = false;
    }

    int maxCount = separatorsCount.Max();

    return maxCount == 0 ? '\0' : separators[separatorsCount.IndexOf(maxCount)];
}


A CSV stream is represented with a reader parameter that is used for reading characters from a CSV stream; parameter rowCount tells the method how many rows should be read before determining the separator and the separators parameter is a list of characters that tells the method which characters are possible separators.

Method maintains internal state with these parameters:

  • separatorsCount - used for counting the number of occurrences of a possible separator as a separator in a CSV stream,

  • character - last character that was read from the CSV stream,

  • row - index of the currently processing row in the CSV stream,

  • quoted - true if characters that are read next are enclosed in quotes, otherwise false,

  • firstChar - true if next character that is to be read is the first character of the next entry in the CSV stream. This parameter is needed because we consider a value to be enclosed in quotes only if the opening quote is the first character of the CSV entry.


When rowCount rows are read or a CSV stream is read to the end, the method returns the first of the possible separators that has a maximum number of occurrences as a separator in the CSV stream. If any of the possible separators never occurred as a separator in CSV stream, '\0' is returned.
The method takes care when reading quotes, separators and new line characters that are part of the quoted value. In this case, if a quote is read, the method will peek into the CSV stream to see if the next character is also a quote, otherwise it will consider this quote to be a closing quote. New line and separator characters are ignored if contained in a quoted value.

For example, in the following Employees.csv file:

Name,Surname,Salary
John,Doe,"$2,130"
Fred;Nurk;"$1,500"
Hans;Meier;"$1,650"
Ivan;Horvat;"$3,200"

The method detects that the CSV separator is (";") although the total number of occurrences of (";") is 6 and the total number of occurrences of (",") is 8. That is because the last 4 occurrences of (",") are enclosed in quotes so they don't qualify as possible separators. So the total number of occurrences of (",") as separators is 4 and the total number of occurrences of (";") as separators is 6, which makes (";") the most probable CSV separator.

Bundled with this article is a WPF solution that demonstrates auto detection of the CSV separator in action. Solution can be downloaded here. The application is located in the bin/Release folder. Original article can be viewed here.

Alternative

It shouldn't be too hard to derive an entire CSV reader from the code presented in this article, but tabular data can come in many different formats and implementing a reader and a writer for each of them may not be so easy and could really hurt your productivity.

For that reason, you could use a third party component which supports various file formats. This will probably cost you some money, but formats like XLS, XLSX, CSV, ODS, HTML are likely to be supported within the same API, so your application will be able to target different file formats using the same code.
 


Similar Articles