Csv::Parser (csv-parser)

Compile-time and runtime CSV parser written in Modern C++

Features

• Header-only.
• Requires only standard C++ (minimum C++17, with C++23 support providing additional functionality).
• Supports reading CSV data from std::string_view at compile-time.
• Fully supports RFC 4180, including quoted values, escaped quotes, and newlines in field values.
• Liberal in terms of accepting not-quite-standard CSV files, but detects errors when needed.
• Supports Excel CSV variations.
• Supports reading data as different types (string, double, empty field), with some restrictions when parsing at compile-time.
• Modular design allows for easy extension of the library.
• Extensively tested using Catch2.

API Reference

Automatically generated API reference describes the public API in detail.

Usage Examples

Runtime Parsing into 2D std::vector

Example:

#include "csv_parser.h"
 
// ...
 
using namespace std::string_view_literals;
 
// Data to parse
std::string_view data = "abc,def\n5,6"sv;
 
// Let "cell_refs" be a vector of columns.
// After parsing, each element will contain Csv::CellReference object. If the cell data type
// is Csv::CellType::String, Csv::CellReference object will reference a part of the original data.
// Other Cell* types, as well as floating point and integral types can also be used here.
std::vector<std::vector<Csv::CellReference>> cell_refs;
 
Csv::Parser parser;
 
try {
    // This throws Csv::ParseError on error.
    parser.parseTo2DVector(data, cell_refs);
}
catch(Csv::ParseError& ex) {
    std::cerr << "CSV parse error: " << ex.what() << std::endl;
    return EXIT_FAILURE;
}
 
assert(cell_refs.size() == 2);
assert(cell_refs[0].size() == 2);
assert(cell_refs[1].size() == 2);
 
assert(cell_refs[0][0].getType() == Csv::CellType::String);
assert(cell_refs[1][0].getType() == Csv::CellType::String);
assert(cell_refs[0][1].getType() == Csv::CellType::Double);
assert(cell_refs[1][1].getType() == Csv::CellType::Double);
 
std::cout << "Column 0, row 0: " << cell_refs[0][0].getCleanString().value() << std::endl;  // abc
std::cout << "Column 1, row 0: " << cell_refs[1][0].getCleanString().value() << std::endl;  // def
std::cout << "Column 0, row 1: " << cell_refs[0][1].getDouble().value() << std::endl;  // 5
std::cout << "Column 1, row 1: " << cell_refs[1][1].getDouble().value() << std::endl;  // 6

Runtime Parsing of Numeric Matrix Into 1D Vector With Row-Major Order

Example:

#include "csv_parser.h"
 
// ...
 
using namespace std::string_view_literals;
 
// Data to parse
std::string_view data = "11,12,13\n21,22,23"sv;
 
// Let matrix_data be a flat matrix of doubles in row-major order.
// Other floating point and integral types, as well as Cell* types can also be used here.
std::vector<double> matrix_data;
 
Csv::Parser parser;
Csv::MatrixInformation info;
 
try {
    // This throws Csv::ParseError on error.
    info = parser.parseToVectorRowMajor(data, matrix_data);
}
catch(Csv::ParseError& ex) {
    std::cerr << "CSV parse error: " << ex.what() << std::endl;
    return EXIT_FAILURE;
}
 
assert(matrix_data.size() == 3 * 2);
assert(info.getColumns() == 3);
assert(info.getRows() == 2);
 
std::cout << "Row 0, column 0: " << matrix_data[0] << std::endl;  // 11
std::cout << "Row 0, column 1: " << matrix_data[1] << std::endl;  // 12
std::cout << "Row 0, column 2: " << matrix_data[2] << std::endl;  // 13
 
// matrixIndex(row, column) can be used to avoid accidental mistakes
std::cout << "Row 1, column 0: " << matrix_data[info.matrixIndex(1, 0)] << std::endl;  // 21
std::cout << "Row 1, column 1: " << matrix_data[info.matrixIndex(1, 1)] << std::endl;  // 22
std::cout << "Row 1, column 2: " << matrix_data[info.matrixIndex(1, 2)] << std::endl;  // 23
 
return EXIT_SUCCESS;

Compile-Time Parsing Into 2D std::array

Currently, parsing at compile-time has some restrictions:

• To collapse consecutive double-quotes in strings, a compile-time-allocated buffer has to be used.
• By default, only string_views are supported for output (no numeric types).
  • If using C++23 or later, integral types are supported as well.
  • If compile-time parsing of floating point numbers is needed, fast_float can be plugged in.

One (possibly useful) consequence of compile-time parsing is that a parse error also causes a compilation error.

Example:

#include "csv_parser.h"
 
// ...
 
using namespace std::string_view_literals;
 
constexpr std::size_t columns = 2, rows = 2;
constexpr std::string_view data =
R"(abc, "def"
"with ""quote inside",6)";
 
constexpr Csv::Parser parser;
 
// parse into std::array<std::array<CellStringReference, rows>, columns>
constexpr auto matrix = parser.parseTo2DArray<rows, columns>(data);
 
// Verify the data at compile time.
// Note that consecutive double-quotes are not collapsed when using
// getOriginalStringView(). To collapse them, use the getCleanStringBuffer()
// approach below.
static_assert(matrix[0][0].getOriginalStringView() == "abc"sv);
static_assert(matrix[1][0].getOriginalStringView() == "def"sv);
static_assert(matrix[1][1].getOriginalStringView() == "6"sv);
 
// To support consecutive double-quote collapsing at compile-time, allocate a compile-time
// buffer to place the clean string inside. The buffer size has to be at least that
// of a collapsed string value.
// If the buffer is too small, the code will simply not compile.
constexpr auto buffer_size = matrix[0][1].getRequiredBufferSize();  // collapsed size
constexpr auto buffer = matrix[0][1].getCleanStringBuffer<buffer_size>();
static_assert(buffer.getStringView() == R"(with "quote inside)"sv);

Compile-Time Parsing of Integral Matrix Into 1D Vector

The library can be used to parse CSV data at compile-time into a 1D vector, in row-major or column-major order. Additionally, compile-time parsing of integral types is supported since C++23.

Example:

#include "csv_parser.h"
 
// ...
 
using namespace std::string_view_literals;
 
constexpr std::size_t columns = 2, rows = 3;
constexpr std::string_view data =
R"(11, -12
21, 4
60, -10)";
 
// Use Csv::LocaleUnawareBehaviorPolicy to avoid locale issues and allow compile-time integer parsing.
constexpr Csv::Parser<Csv::LocaleUnawareBehaviorPolicy> parser;
 
// Parse into std::array<CellStringReference, rows * columns> in row-major order
{
    constexpr auto matrix = parser.parseToArray<rows, columns>(data, Csv::MatrixOrder::RowMajor);
 
    static_assert(matrix[0].getOriginalStringView() == "11"sv);
    static_assert(matrix[2].getOriginalStringView() == "21"sv);
}
 
// Parse into std::array<std::int64_t, rows * columns> in column-major order.
// Compile-time parsing to integers is supported since C++23.
#if __cplusplus >= 202300L
{
    constexpr auto matrix = parser.parseToArray<rows, columns, std::int64_t>(data, Csv::MatrixOrder::ColumnMajor);
 
    static_assert(matrix[0] == 11);
    static_assert(matrix[2] == 60);
}
#endif

Compile-Time Parsing of Numeric Matrix Using External Library for Floating Point Numbers

A library like fast_float can be used to parse floating point numbers at compile-time. This is done by creating a custom policy that implements the readNumber and create methods. This approach also allows for improving the performance of parsing floating point numbers compared to the standard library.

Example:

#include "csv_parser.h"
#include "fast_float/fast_float.h"
 
// Custom policy which uses fast_float library for parsing to floating point.
struct CustomPolicy : Csv::LocaleUnawareBehaviorPolicy {
    template<typename Number>
    static constexpr std::optional<Number> readNumber(std::string_view cell)
    {
        Number parsed_value = 0;
        auto [ptr, ec] = fast_float::from_chars(cell.begin(), cell.end(), parsed_value);
        if (ec == std::errc() && ptr == (cell.end())) {
            return parsed_value;
        }
        return std::nullopt;
    }
 
    template<typename CellT>
    static constexpr CellT create(std::string_view cell, Csv::CellTypeHint hint)
    {
        return readNumber<CellT>(cell).value_or(std::numeric_limits<CellT>::quiet_NaN());
    }
};
 
 
void parse()
{
    constexpr std::string_view data =
    R"(11.6, -12.3
2e5, -inf
nan, inf)";
 
    // Use custom policy which uses fast_float library for parsing to floating point.
    constexpr Csv::Parser<CustomPolicy> parser;
    constexpr std::size_t columns = 2, rows = 3;
 
    // Parse into std::array<double, rows * columns> in row-major order
    constexpr auto matrix = parser.parseToArray<rows, columns, double>(data, Csv::MatrixOrder::RowMajor);
 
    static_assert(matrix[0] == 11.6);
    static_assert(matrix[3] == -std::numeric_limits<double>::infinity());
}

Copyright

Copyright: Alexander Shaduri ashad.nosp@m.uri@.nosp@m.gmail.nosp@m..com
License: Zlib