From Free Pascal wiki
Jump to navigationJump to search

English (en) русский (ru)


CSV means Comma-Separated Values, and is a popular file format that is unfortunately not completely standardized.

It is a text based file format with

  • data fields separated by commas (or, in variants, other characters like tabs and semicolons)
  • there may or not be a header line that lists the field names
  • field data containing delimiters may be prohibited or enclosed by quotes (most commonly the double quote character)
  • line endings (#13 and/or #10) may or may not be allowed in field data

RFC4180 tries to codify and standardize existing practice; it makes sense to conform to this standard when writing CSV data (and accept all RFC4180 data when reading).

A different, Creativyst/Excel, CSV pseudo-format specification can be found here.

A sample CSV snippet:

Jim,Weston,19560818,"Also known as ""The Butcher"""

Apparently a header line is used here, as is quoting using double quotes.

Spreadsheet packages such as Microsoft Excel and OpenOffice/LibreOffice Calc are able to export to and import from this format. However, as Microsoft Excel may interpret some fields such as date fields differently depending on a user's OS locale, it may pay to find alternative ways of transferring data (e.g. using the FPSpreadsheet code).


Delphi (and Free Pascal) have a very similar (but not identical) format, SDF. See SDF for more details.



CsvDocument is a robust implementation of both the RFC 4180 CSV format and the alternative Creativyst/Excel CSV format. It offers both line-based and document-based access. Recommended for use with FPC/Lazarus. See CsvDocument.


TStringList offers the DelimitedText property. This parses a line of text out into separate fields. Note, however, that DelimitedText is supposed to be in SDF, a Delphi-specific format that is very much like CSV, but does not conform with RFC4180 completely.

Tips: when reading CSV data, set the StrictDelimiter property to true.

When writing out CSV data, set StrictDelimiter to false and output the DelimitedText property. One oddity is that e.g. tab characters are removed when writing out data using StrictDelimiter:=false

SDF format

See SDF format


Free Pascal offers the TSdfDataSet, which stores data in SDF format. Note: FPC 2.6.x and earlier store sdfdataset data in a format that is not completely CSV or completely SDF compatible.

As indicated, SDF differs from CSV. Depending on the flavour of CSV, this format may be close enough to what a reading application expects to function. If you need an RFC4180 compliant CSV format, use TCSVDataSet instead.

TSdfDataset and TFixedDataset sample files

Sample SDF file

Below is a sample database for TSdfDataset. Note that the first line has the names of the fields and that we are using commas as separators:
3,transistor npn,transistor npn
  • Sample TFixedDataset file
Each record occupies a fixed amount of space, and if the field is smaller, spaces should be used to fill the remaining size.
Name = 15 chars; Surname = 15 chars; e_mail = 20 chars;

Piet Pompies piet@pompies.net

Using the datasets: example code

Sometimes it is useful to create the dataset and work with it completely in code, and the following code will do exactly this. Note some peculiarities of TSdfDataset/TFixedDataset:

  • The lines in the database can have a maximum size of about 300. A fix is being researched.
  • It is necessary to add the field definitions. Some datasets are able to fill this information alone from the database file.
  • One should set FirstLineAsSchema to true, to indicate that the first line includes the field names and positions.
  • The Delimiter property holds the separator for the fields. It will not be possible to use this char in strings in the database. Similarly it will currently not be possible to have line endings in the database because they mark the change between records. It's possible to overcome this by substituting the needed comma or line ending with another rarely used char, like #. When showing the data on screen all # chars could be converted to line endings and the inverse when storing data back to the database. The ReplaceString routine is useful here.
uses sdfdata, db;

constructor TComponentsDatabase.Create;
  FDataset: TSdfDataset;
  inherited Create;

  FDataset := TSdfDataset.Create(nil);
  FDataset.FileName := vConfigurations.ComponentsDBFile;
  FDataset.FileMustExist := false; //don't require existing csv file

  // Not necessary with TSdfDataset:
  //  FDataset.TableName := STR_DB_COMPONENTS_TABLE;
  //  FDataset.PrimaryKey := STR_DB_COMPONENTS_ID;

  // Adds field definitions
  FDataset.FieldDefs.Add('ID', ftString);
  //schema addition is needed so that sdfdataset writes out the csv header correctly:
  FDataset.FieldDefs.Add('NAMEEN', ftString);
  FDataset.FieldDefs.Add('NAMEPT', ftString);
  FDataset.FieldDefs.Add('HEIGHT', ftString);
  FDataset.FieldDefs.Add('WIDTH', ftString);
  FDataset.FieldDefs.Add('PINS', ftString);
  FDataset.FieldDefs.Add('DRAWINGCODE', ftString);

  // Necessary for TSdfDataset
  FDataset.Delimiter := ',';
  FDataset.FirstLineAsSchema := True;

  FDataset.Active := True;

  // Sets the initial record
  CurrentRecNo := 1;

When using TSdfDataSet directly be aware that RecNo, although it is implemented, does not work as a way to move through the dataset whether reading or writing records. The standard navigation routines like First, Next, Prior and Last work as expected, so you need to use them rather than RecNo.

If you are used to using absolute record numbers to navigate around a database you can implement your own version of RecNo. Declare a global longint variable called CurrentRecNo which will hold the current RecNo value. Remember that this variable will have the same convention as RecNo, so the first record has number 1 (it is not zero-based). After activating the database initialize the database to the first record with TSdfDataset.First and set CurrentRecNo := 1.

Example code:

  Moves to the desired record using TDataset.Next and TDataset.Prior
  This avoids using TDataset.RecNo which doesn't navigate reliably in any dataset.

  @param AID Indicates the record number. The first record has number 1
procedure TComponentsDatabase.GoToRec(AID: Integer);
  // We are before the desired record, move forward
  if CurrentRecNo < AID then
    while (not FDataset.EOF) and (CurrentRecNo < AID) do
  // We are after the desired record, move back
  else if CurrentRecNo > AID  then
    while (CurrentRecNo >= 1) and (CurrentRecNo > AID) do

Data Export

Free Pascal/Lazarus database export functionality (e.g. TCSVExporter on the Data Export tab - requires installation of the lazdbexport package) offers CSV export functionality for datasets.

Jan's CSV components

See JCSV (Jans CSV Components).


ZMSQL stores data in semicolon-delimited files (using SDF?).