A.10 library(csv): Process CSV (Comma-Separated Values) data

See also

RFC 4180

To be done

- Implement immediate assert of the data to avoid possible stack overflows.
- Writing creates an intermediate code-list, possibly overflowing resources. This waits for pure output!

This library parses and generates CSV data. CSV data is represented in Prolog as a list of rows. Each row is a compound term, where all rows have the same name and arity.

[det]csv_read_file(+File, -Rows)

[det]csv_read_file(+File, -Rows, +Options)

Read a CSV file into a list of rows. Each row is a Prolog term with the same arity. Options is handed to csv/4. Remaining options are processed by phrase_from_file/3. The default separator depends on the file name extension and is \t for .tsv files and , otherwise.

Suppose we want to create a predicate table/6 from a CSV file that we know contains 6 fields per record. This can be done using the code below. Without the option arity(6), this would generate a predicate table/N, where N is the number of fields per record in the data.

?- csv_read_file(File, Rows, [functor(table), arity(6)]),
   maplist(assert, Rows).

[det]csv(?Rows)//

[det]csv(?Rows, +Options)//

Prolog DCG to `read/write' CSV data. Options:

separator(+Code)

The comma-separator. Must be a character code. Default is (of course) the comma. Character codes can be specified using the 0' notion. E.g., using separator(0';) parses a semicolon separated file.

ignore_quotes(+Boolean)

If true (default false), threat double quotes as a normal character.

strip(+Boolean)

If true (default false), strip leading and trailing blank space. RFC4180 says that blank space is part of the data.

convert(+Boolean)

If true (default), use name/2 on the field data. This translates the field into a number if possible.

functor(+Atom)

Functor to use for creating row terms. Default is row.

arity(?Arity)

Number of fields in each row. This predicate raises a domain_error(row_arity(Expected), Found) if a row is found with different arity.

match_arity(+Boolean)

If false (default true), do not reject CSV files where lines provide a varying number of fields (columns). This can be a work-around to use some incorrect CSV files.

[nondet]csv_read_file_row(+File, -Row, +Options)

True when Row is a row in File. First unifies Row with the first row in File. Backtracking yields the second, ... row. This interface is an alternative to csv_read_file/3 that avoids loading all rows in memory. Note that this interface does not guarantee that all rows in File have the same arity.

In addition to the options of csv_read_file/3, this predicate processes the option:

line(-Line)

Line is unified with the 1-based line-number from which Row is read. Note that Line is not the physical line, but rather the logical record number.

To be done

Input is read line by line. If a record separator is embedded in a quoted field, parsing the record fails and another line is added to the input. This does not nicely deal with other reasons why parsing the row may fail.

[det]csv_write_file(+File, +Data)

[det]csv_write_file(+File, +Data, +Options)

Write a list of Prolog terms to a CSV file. Options are given to csv/4. Remaining options are given to open/4. The default separator depends on the file name extension and is \t for .tsv files and , otherwise.

[det]csv_write_stream(+Stream, +Data, +Options)

Write the rows in Data to Stream. This is similar to csv_write_file/3, but can deal with data that is produced incrementally. The example below saves all answers from the predicate data/3 to File.

save_data(File) :-
   setup_call_cleanup(
       open(File, write, Out),
       forall(data(C1,C2,C3),
              csv_write_stream(Out, [row(C1,C2,C3)], [])),
       close(Out)),