[DEPRECATED APRIL 2020]

This library is now deprecated. Checkout a second implementation of this library here: https://github.com/p-ranav/csv2.

Highlights

• Header-only library
• Fast, asynchronous, multi-threaded processing using:
  • Lock-free Concurrent Queues
  • Robin hood Hashing
• Requires C++17
• MIT License

Table of Contents

• Reading CSV files
  • Dialects
    • Configuring Custom Dialects
  • Multi-character Delimiters
  • Ignoring Columns
  • No Header?
  • Dealing with Empty Rows
  • Reading first N rows
  • Performance Benchmark
• Writing CSV files
• Steps For Contributors
• Steps For Users
• Continuous Integration Reports
• License

Reading CSV files

Simply include reader.hpp and you're good to go.

#include <csv/reader.hpp>

To start parsing CSV files, create a csv::Reader object and call .read(filename).

csv::Reader foo;
foo.read("test.csv");

This .read method is non-blocking. The reader spawns multiple threads to tokenize the file stream and build a "list of dictionaries". While the reader is doing it's thing, you can start post-processing the rows it has parsed so far using this iterator pattern:

while(foo.busy()) {
  if (foo.ready()) {
    auto row = foo.next_row();  // Each row is a csv::unordered_flat_map (github.com/martinus/robin-hood-hashing)
    auto foo = row["foo"]       // You can use it just like an std::unordered_map
    auto bar = row["bar"];
    // do something
  }
}

If instead you'd like to wait for all the rows to get processed, you can call .rows() which is a convenience method that executes the above while loop

auto rows = foo.rows();           // blocks until the CSV is fully processed
for (auto& row : rows) {          // Example: [{"foo": "1", "bar": "2"}, {"foo": "3", "bar": "4"}, ...] 
  auto foo = row["foo"];
  // do something
}

Dialects

This csv library comes with three standard dialects:

Name	Description
excel	The excel dialect defines the usual properties of an Excel-generated CSV file
excel_tab	The excel_tab dialect defines the usual properties of an Excel-generated TAB-delimited file
unix	The unix dialect defines the usual properties of a CSV file generated on UNIX systems, i.e. using '\n' as line terminator and quoting all fields

Configuring Custom Dialects

Custom dialects can be constructed with .configure_dialect(...)

csv::Reader csv;
csv.configure_dialect("my fancy dialect")
  .delimiter("")
  .quote_character('"')
  .double_quote(true)
  .skip_initial_space(false)
  .trim_characters(' ', '\t')
  .ignore_columns("foo", "bar")
  .header(true)
  .skip_empty_rows(true);

csv.read("foo.csv");
for (auto& row : csv.rows()) {
  // do something
}

Property	Data Type	Description
delimiter	std::string	specifies the character sequence which should separate fields (aka columns). Default = ","
quote_character	char	specifies a one-character string to use as the quoting character. Default = '"'
double_quote	bool	controls the handling of quotes inside fields. If true, two consecutive quotes should be interpreted as one. Default = true
skip_initial_space	bool	specifies how to interpret whitespace which immediately follows a delimiter; if false, it means that whitespace immediately after a delimiter should be treated as part of the following field. Default = false
trim_characters	std::vector<char>	specifies the list of characters to trim from every value in the CSV. Default = {} - nothing trimmed
ignore_columns	std::vector<std::string>	specifies the list of columns to ignore. These columns will be stripped during the parsing process. Default = {} - no column ignored
header	bool	indicates whether the file includes a header row. If true the first row in the file is a header row, not data. Default = true
column_names	std::vector<std::string>	specifies the list of column names. This is useful when the first row of the CSV isn't a header Default = {}
skip_empty_rows	bool	specifies how empty rows should be interpreted. If this is set to true, empty rows are skipped. Default = false

The line terminator is '\n' by default. I use std::getline and handle stripping out '\r' from line endings. So, for now, this is not configurable in custom dialects.

Multi-character Delimiters

Consider this strange, messed up log file:

[Thread ID] :: [Log Level] :: [Log Message] :: {Timestamp}
04 :: INFO :: Hello World ::             1555164718
02        :: DEBUG :: Warning! Foo has happened                :: 1555463132

To parse this file, simply configure a new dialect that splits on "::" and trims whitespace, braces, and bracket characters.

csv::Reader csv;
csv.configure_dialect("my strange dialect")
  .delimiter("::")
  .trim_characters(' ', '[', ']', '{', '}');   

csv.read("test.csv");
for (auto& row : csv.rows()) {
  auto thread_id = row["Thread ID"];    // "04"
  auto log_level = row["Log Level"];    // "INFO"
  auto message = row["Log Message"];    // "Hello World"
  // do something
}

Ignoring Columns

Consider the following CSV. Let's say you don't care about the columns age and gender. Here, you can use .ignore_columns and provide a list of columns to ignore.

name, age, gender, email, department
Mark Johnson, 50, M, mark.johnson@gmail.com, BA
John Stevenson, 35, M, john.stevenson@gmail.com, IT
Jane Barkley, 25, F, jane.barkley@gmail.com, MGT

You can configure the dialect to ignore these columns like so:

csv::Reader csv;
csv.configure_dialect("ignore meh and fez")
  .delimiter(", ")
  .ignore_columns("age", "gender");

csv.read("test.csv");
auto rows = csv.rows();
// Your rows are:
// [{"name": "Mark Johnson", "email": "mark.johnson@gmail.com", "department": "BA"},
//  {"name": "John Stevenson", "email": "john.stevenson@gmail.com", "department": "IT"},
//  {"name": "Jane Barkley", "email": "jane.barkley@gmail.com", "department": "MGT"}]

No Header?

Sometimes you have CSV files with no header row:

9 52 1
52 91 0
91 135 0
135 174 0
174 218 0
218 260 0
260 301 0
301 341 0
341 383 0
...

If you want to prevent the reader from parsing the first row as a header, simply:

• Set .header to false
• Provide a list of column names with .column_names(...)

csv.configure_dialect("no headers")
  .header(false)
  .column_names("foo", "bar", "baz");

The CSV rows will now look like this:

[{"foo": "9", "bar": "52", "baz": "1"}, {"foo": "52", "bar": "91", "baz": "0"}, ...]

If .column_names is not called, then the reader simply generates dictionary keys like so:

[{"0": "9", "1": "52", "2": "1"}, {"0": "52", "1": "91", "2": "0"}, ...]

Dealing with Empty Rows

Sometimes you have to deal with a CSV file that has empty lines; either in the middle or at the end of the file:

a,b,c
1,2,3

4,5,6

10,11,12

Here's how this get's parsed by default:

csv::Reader csv;
csv.read("inputs/empty_lines.csv");
auto rows = csv.rows();
// [{"a": 1, "b": 2, "c": 3}, {"a": "", "b": "", "c": ""}, {"a": "4", "b": "5", "c": "6"}, {"a": "", ...}]

If you don't care for these empty rows, simply call .skip_empty_rows(true)

csv::Reader csv;
csv.configure_dialect()
  .skip_empty_rows(true);
csv.read("inputs/empty_lines.csv");
auto rows = csv.rows();
// [{"a": 1, "b": 2, "c": 3}, {"a": "4", "b": "5", "c": "6"}, {"a": "10", "b": "11", "c": "12"}]

Reading first N rows

If you know exactly how many rows to parse, you can help out the reader by using the .read(filename, num_rows) overloaded method. This saves the reader from trying to figure out the number of lines in the CSV file. You can use this method to parse the first N rows of the file instead of parsing all of it.

csv::Reader foo;
foo.read("bar.csv", 1000);
auto rows = foo.rows();

Note: Do not provide num_rows greater than the actual number of rows in the file - The reader will loop forever till the end of time.

Performance Benchmark

// benchmark.cpp
void parse(const std::string& filename) {
  csv::Reader foo;
  foo.read(filename);
  std::vector<csv::unordered_flat_map<std::string_view, std::string>> rows;
  while (foo.busy()) {
    if (foo.ready()) {
      auto row = foo.next_row();
      rows.push_back(row);
    }
  }
}

$ g++ -pthread -std=c++17 -O3 -Iinclude/ -o test benchmark.cpp
$ time ./test

Each test is run 30 times on an Intel(R) Core(TM) i7-6650-U @ 2.20 GHz CPU.

Here are the average-case execution times:

Dataset	File Size	Rows	Cols	Time
Demographic Statistics By Zip Code	27 KB	237	46	0.026s
Simple 3-column CSV	14.1 MB	761,817	3	0.523s
Majestic Million	77.7 MB	1,000,000	12	1.972s
Crimes 2001 - Present	1.50 GB	6,846,406	22	32.411s

Writing CSV files

Simply include writer.hpp and you're good to go.

#include <csv/writer.hpp>

To start writing CSV files, create a csv::Writer object and provide a filename:

csv::Writer foo("test.csv");

Constructing a writer spawns a worker thread that is ready to start writing rows. Using .configure_dialect, configure the dialect to be used by the writer. This is where you can specify the column names:

foo.configure_dialect()
  .delimiter(", ")
  .column_names("a", "b", "c");

Now it's time to write rows. You can do this in multiple ways:

foo.write_row("1", "2", "3");                                     // parameter packing
foo.write_row({"4", "5", "6"});                                   // std::vector
foo.write_row(std::map<std::string, std::string>{                 // std::map
  {"a", "7"}, {"b", "8"}, {"c", "9"} });
foo.write_row(std::unordered_map<std::string, std::string>{       // std::unordered_map
  {"a", "7"}, {"b", "8"}, {"c", "9"} });
foo.write_row(csv::unordered_flat_map<std::string, std::string>{  // csv::unordered_flat_map
  {"a", "7"}, {"b", "8"}, {"c", "9"} });

You can also omit one or more values dynamically when using maps:

foo.write_row(std::map<std::string, std::string>{                 // std::map
  {"a", "7"}, {"c", "9"} });                                      // omitting "b"
foo.write_row(std::unordered_map<std::string, std::string>{       // std::unordered_map
  {"b", "8"}, {"c", "9"} });                                      // omitting "a"
foo.write_row(csv::unordered_flat_map<std::string, std::string>{  // csv::unordered_flat_map
  {"a", "7"}, {"b", "8"} });                                      // omitting "c"

Finally, once you're done writing rows, call .close() to stop the worker thread and close the file stream.

foo.close();

Here's an example writing 3 million lines of CSV to a file:

csv::Writer foo("test.csv");
foo.configure_dialect()
  .delimiter(", ")
  .column_names("a", "b", "c");

for (long i = 0; i < 3000000; i++) {
  auto x = std::to_string(i % 100);
  auto y = std::to_string((i + 1) % 100);
  auto z = std::to_string((i + 2) % 100);
  foo.write_row(x, y, z);
}
foo.close();

The above code takes about 1.8 seconds to execute on my Surface Pro 4.

Steps For Contributors

Contributions are welcome, have a look at the CONTRIBUTING.md document for more information.

git clone https://github.com/p-ranav/csv.git
cd csv
git submodule update --init --recursive
mkdir build
cd build
cmake .. -DCSV_BUILD_TESTS=ON
cmake --build . --config Debug
ctest --output-on-failure -C Debug

Steps For Users

git clone https://github.com/p-ranav/csv.git
cd csv
mkdir build
cd build
cmake ../.
sudo make install

Continuous Integration Reports

• AppVeyor
• Travis CI

License

The project is available under the MIT license.