ftcsv

ftcsv is a fast csv library written in pure Lua. It's been tested with LuaJIT 2.0/2.1 and Lua 5.1, 5.2, 5.3, and 5.4

It features two parsing modes, one for CSVs that can easily be loaded into memory (up to a few hundred MBs depending on the system), and another for loading files using an iterator - useful for manipulating large files or processing during load. It correctly handles most csv (and csv-like) files found in the wild, from varying line endings (Windows, Linux, and OS9), UTF-8 BOM support, and odd delimiters. There are also various options that can tweak how a file is loaded, only grabbing a few fields, renaming fields, and parsing header-less files!

Installing

You can either grab ftcsv.lua from here or install via luarocks:

luarocks install ftcsv

Parsing

There are two main parsing methods: ftcv.parse and ftcsv.parseLine. ftcsv.parse loads the entire file and parses it, while ftcsv.parseLine is an iterator that parses one line at a time.

ftcsv.parse(fileName, delimiter [, options])

ftcsv.parse will load the entire csv file into memory, then parse it in one go, returning a lua table with the parsed data and a lua table containing the column headers. It has only two required parameters - a file name and delimiter (limited to one character). A few optional parameters can be passed in via a table (examples below).

Just loading a csv file:

local ftcsv = require('ftcsv')
local zipcodes, headers = ftcsv.parse("free-zipcode-database.csv", ",")

ftcsv.parseLine(fileName, delimiter, [, options])

ftcsv.parseLine will open a file and read options.bufferSize bytes of the file. bufferSize defaults to 2^16 bytes (which provides the fastest parsing on most unix-based systems), or can be specified in the options. ftcsv.parseLine is an iterator and returns one line at a time. When all the lines in the buffer are read, it will read in another bufferSize bytes of a file and repeat the process until the entire file has been read.

If specifying bufferSize there are a couple of things to remember:

• bufferSize must be at least the length of the longest row.
• If bufferSize is too small, an error is returned.
• If bufferSize is the length of the entire file, all of it will be read and returned one line at a time (performance is roughly the same as ftcsv.parse).

Parsing through a csv file:

local ftcsv = require("ftcsv")
for zipcode in ftcsv.parseLine("free-zipcode-database.csv", ",") do
    print(zipcode.Zipcode)
    print(zipcode.State)
end

Options

The options are the same for parseLine and parse, with the exception of loadFromString and bufferSize. loadFromString only works with parse and bufferSize can only be specified for parseLine.

The following are optional parameters passed in via the third argument as a table.

• loadFromString

  If you want to load a csv from a string instead of a file, set loadFromString to true (default: false)

  ftcsv.parse("a,b,c\r\n1,2,3", ",", {loadFromString=true})
  

• rename

  If you want to rename a field, you can set rename to change the field names. The below example will change the headers from a,b,c to d,e,f

  Note: You can rename two fields to the same value, ftcsv will keep the field that appears latest in the line.

  local options = {loadFromString=true, rename={["a"] = "d", ["b"] = "e", ["c"] = "f"}}
  local actual = ftcsv.parse("a,b,c\r\napple,banana,carrot", ",", options)
  

• fieldsToKeep

  If you only want to keep certain fields from the CSV, send them in as a table-list and it should parse a little faster and use less memory.

  Note: If you want to keep a renamed field, put the new name of the field in fieldsToKeep:

  local options = {loadFromString=true, fieldsToKeep={"a","f"}, rename={["c"] = "f"}}
  local actual = ftcsv.parse("a,b,c\r\napple,banana,carrot\r\n", ",", options)
  

  Also Note: If you apply a function to the headers via headerFunc, and want to select fields from fieldsToKeep, you need to have what the post-modified header would be in fieldsToKeep.

• ignoreQuotes

  If ignoreQuotes is true, it will leave all quotes in the final parsed output. This is useful in situations where the fields aren't quoted, but contain quotes, or if the CSV didn't handle quotes correctly and you're trying to parse it.

  local options = {loadFromString=true, ignoreQuotes=true}
  local actual = ftcsv.parse('a,b,c\n"apple,banana,carrot', ",", options)
  

• headerFunc

  Applies a function to every field in the header. If you are using rename, the function is applied after the rename.

  Ex: making all fields uppercase

  local options = {loadFromString=true, headerFunc=string.upper}
  local actual = ftcsv.parse("a,b,c\napple,banana,carrot", ",", options)
  

• headers

  Set headers to false if the file you are reading doesn't have any headers. This will cause ftcsv to create indexed tables rather than a key-value tables for the output.

  local options = {loadFromString=true, headers=false}
  local actual = ftcsv.parse("apple>banana>carrot\ndiamond>emerald>pearl", ">", options)
  

  Note: Header-less files can still use the rename option and after a field has been renamed, it can specified as a field to keep. The rename syntax changes a little bit:

  local options = {loadFromString=true, headers=false, rename={"a","b","c"}, fieldsToKeep={"a","b"}}
  local actual = ftcsv.parse("apple>banana>carrot\ndiamond>emerald>pearl", ">", options)
  

  In the above example, the first field becomes 'a', the second field becomes 'b' and so on.

For all tested examples, take a look in /spec/feature_spec.lua

The options can be string together. For example if you wanted to loadFromString and not use headers, you could use the following:

ftcsv.parse("apple,banana,carrot", ",", {loadFromString=true, headers=false})

Encoding

ftcsv.encode(inputTable, delimiter[, options])

ftcsv.encode takes in a lua table and turns it into a text string that can be written to a file. It has two required parameters, an inputTable and a delimiter. You can use it to write out a file like this:

local fileOutput = ftcsv.encode(users, ",")
local file = assert(io.open("ALLUSERS.csv", "w"))
file:write(fileOutput)
file:close()

Options

• fieldsToKeep

  if fieldsToKeep is set in the encode process, only the fields specified will be written out to a file.

  local output = ftcsv.encode(everyUser, ",", {fieldsToKeep={"Name", "Phone", "City"}})
  

• onlyRequiredQuotes

  if onlyRequiredQuotes is set to true, the output will only include quotes around fields that are quotes, have newlines, or contain the delimter.

  local output = ftcsv.encode(everyUser, ",", {noQuotes=true})
  

Error Handling

ftcsv returns a litany of errors when passed a bad csv file or incorrect parameters. You can find a more detailed explanation of the more cryptic errors in ERRORS.md

Benchmarks

We ran ftcsv against a few different csv parsers (PIL/csvutils, lua_csv, and lpeg_josh) for lua and here is what we found:

20 MB file, every field is double quoted

Parser	Lua	LuaJIT
PIL/csvutils	1.754 +/- 0.136 SD	1.012 +/- 0.112 SD
lua_csv	4.191 +/- 0.128 SD	2.382 +/- 0.133 SD
lpeg_josh	0.996 +/- 0.149 SD	0.725 +/- 0.083 SD
ftcsv	1.342 +/- 0.130 SD	0.301 +/- 0.099 SD

12 MB file, some fields are double quoted

Parser	Lua	LuaJIT
PIL/csvutils	1.456 +/- 0.083 SD	0.691 +/- 0.071 SD
lua_csv	3.738 +/- 0.072 SD	1.997 +/- 0.075 SD
lpeg_josh	0.638 +/- 0.070 SD	0.475 +/- 0.042 SD
ftcsv	1.307 +/- 0.071 SD	0.213 +/- 0.062 SD

LuaCSV was also tried, but usually errored out at odd places during parsing.

NOTE: times are measured using os.clock(), so they are in CPU seconds. Each test was run 30 times in a randomized order. The file was pre-loaded, and only the csv decoding time was measured.

Benchmarks were run under ftcsv 1.2.0

Performance

I did some basic testing and found that in lua, if you want to iterate over a string character-by-character and compare chars, string.byte performs faster than string.sub. As such, ftcsv iterates over the whole file and does byte compares to find quotes and delimiters and then generates a table from it. When using vanilla lua, it proved faster to use string.find instead of iterating character by character (which is faster in LuaJIT), so ftcsv accounts for that and will perform the fastest option that is availble. If you have thoughts on how to improve performance (either big picture or specifically within the code), create a GitHub issue - I'd love to hear about it!

Contributing

Feel free to create a new issue for any bugs you've found or help you need. If you want to contribute back to the project please do the following:

1. If it's a major change (aka more than a quick bugfix), please create an issue so we can discuss it!
2. Fork the repo
3. Create a new branch
4. Push your changes to the branch
5. Run the test suite and make sure it still works
6. Submit a pull request
7. Wait for review
8. Enjoy the changes made!

Licenses

• The main library is licensed under the MIT License. Feel free to use it!
• Some of the test CSVs are from csv-spectrum (BSD-2-Clause) which includes some from csvkit (MIT License)