Originally developed for Wizard, ReadStat is a command-line tool and MIT-licensed C library for reading files from popular stats packages. Supported data formats include:

• SAS: SAS7BDAT (binary file) and XPORT (transport file)
• Stata: DTA (binary file) versions 104-119
• SPSS: POR (portable file), SAV (binary file), and ZSAV (compressed binary)

Supported metadata formats include:

• SAS: SAS7BCAT (catalog file) and .sas (command file)
• Stata: .dct (dictionary file)
• SPSS: .sps (command file)

There is also write support for all the data formats, but not the metadata formats. The produced SAS7BDAT files still cannot be read by SAS, but feel free to contribute your binary-format expertise here.

For reading in R data files, please see the related librdata project.

Grab the latest release and then proceed as usual:

./configure
make
sudo make install

If you're cloning the repository, first make sure you have autotools installed, and then run ./autogen.sh to generate the configure file.

If you're on Mac and see errors about AM_ICONV when you run ./autogen.sh, you'll need to install gettext.

ReadStat now includes a Microsoft Visual Studio project file that includes build targets for the library and tests. See the VS17 folder in the downloaded release for a "one-click" Windows build.

Alternatively, you can build ReadStat on the command line using an msys2 environment. After installing msys2, download some other packages:

pacman -S autoconf automake libtool make mingw-w64-x86_64-toolchain mingw-w64-x86_64-cmake mingw-w64-x86_64-libiconv

Then start a MINGW command line (not the msys2 prompt!) and follow the UNIX install instructions above for this package.

• Julia: ReadStat.jl
• Perl 6: ReadStat.pm6
• Python: pyreadstat
• R: haven

A dockerized version is available here

Standard usage:

readstat [-f] <input file> <output file>

Where:

• <input file> ends with .dta, .por, .sav, .sas7bdat, or .xptand
• <output file> ends with .dta, .por, .sav, .sas7bdat, .xpt or .csv

If libxlsxwriter is found at compile-time, an XLSX file (ending in .xlsx) can be written instead.

If zlib is found at compile-time, compressed SPSS files (.zsav) can be read and written as well.

Use the -f option to overwrite an existing output file.

If you have a plain-text file described by a Stata dictionary file, a SAS command file, or an SPSS command file, a second invocation style is supported:

readstat <input file> <dictionary file> <output file>

Where:

• <input file> can be anything
• <dictionary file> ends with .dct, .sas, or .sps
• <output file> ends with .dta, .por, .sav, .xpt, or .csv

If you have a SAS catalog file containing the data set's value labels, you can use the same invocation:

readstat <input file> <catalog file> <output file>

Except where:

• <input file> ends with .sas7bdat
• <catalog file> ends with .sas7bcat
• <output file> ends with .dta, .por, .sav, .xpt, or .csv

If the file conversion succeeds, ReadStat will report the number of rows and variables converted, e.g.

Converted 111 variables and 160851 rows in 12.36 seconds

At the moment value labels are supported, but the finer nuances of converting format strings (e.g. %8.2g) are not.

A prerequisite for CSV input is that the libcsv library is found at compile time.

CSV input is supported together with a metadata file describing the data:

readstat <input file.csv> <input metadata.json> <output file>

The <output file> should end with .dta, .sav, or .csv.

The <input file.csv> is a regular CSV file.

The <input metadata.json> is a JSON file describing column types, value labels and missing values. The easiest way to create such a metadata file is to use the provided extract_metadata program on an existing file:

$ extract_metadata <input file.(dta|sav|sas7bcat)>

The schema of this JSON file is fully described in variablemetadata_schema.json using JSON Schema.

The following is an example of a valid metadata file:

{
    "type": "SPSS",
    "variables": [
        {
            "type": "NUMERIC",
            "name": "citizenship",
            "label": "Citizenship of respondent",
            "categories": [
                {
                    "code": 1,
                    "label": "Afghanistan"
                },
                {
                    "code": 2,
                    "label": "Albania"
                },
                {
                    "code": 98,
                    "label": "No answer"
                },
                {
                    "code": 99,
                    "label": "Not applicable"
                }
            ],
            "missing": {
                "type": "DISCRETE",
                "values": [
                    98,
                    99
                ]
            }
        }
    ]
}

Here the column citizenship is a numeric column with four possible values 1, 2, 98, and 99. 1 has the label Afghanistan, 2 has Albania, 98 has No answer and 99 has Not applicable. 98 and 99 are defined as missing values.

Other column types are STRING and DATE. All values in DATE columns are expected to conform to ISO 8601 date. Here is an example of DATE metadata:

{
    "type": "SPSS",
    "variables": [
        {
            "type": "DATE",
            "name": "startdate",
            "label": "Start date",
            "categories": [
                {
                    "code": "6666-01-01",
                    "label": "no date available"
                }
            ],
            "missing": {
                "type": "DISCRETE",
                "values": [
                    "6666-01-01",
                    "9999-01-01"
                ]
            }
        }
    ]
}

Value labels are supported for DATE.

The last column type is STRING:

{
    "type": "SPSS",
    "variables": [
        {
            "type": "STRING",
            "name": "somestring",
            "label": "Label of column",
            "missing": {
                "type": "DISCRETE",
                "values": [
                    "NA",
                    "N/A"
                ]
            }
        }
    ]
}

Value labels are not supported for STRING.

The ReadStat API is callback-based. It uses very little memory, and is suitable for programs with progress bars. ReadStat uses iconv to automatically transcode text data into UTF-8, so you don't have to worry about character encodings.

See src/readstat.h for the complete API. In general you'll provide a filename and a set of optional callback functions for handling various information and data found in the file. It's up to the user to store this information in an appropriate data structure. If a context pointer is passed to the parse_* functions, it will be made available to the various callback functions.

Callback functions should return READSTAT_HANDLER_OK (zero) on success. Returning READSTAT_HANDLER_ABORT will abort the parsing process.

Example: Return the number of records in a DTA file.

#include "readstat.h"

int handle_metadata(readstat_metadata_t *metadata, void *ctx) {
    int *my_count = (int *)ctx;

    *my_count = readstat_get_row_count(metadata);

    return READSTAT_HANDLER_OK;
}

int main(int argc, char *argv[]) {
    if (argc != 2) {
        printf("Usage: %s <filename>\n", argv[0]);
        return 1;
    }
    int my_count = 0;
    readstat_error_t error = READSTAT_OK;
    readstat_parser_t *parser = readstat_parser_init();
    readstat_set_metadata_handler(parser, &handle_metadata);

    error = readstat_parse_dta(parser, argv[1], &my_count);

    readstat_parser_free(parser);

    if (error != READSTAT_OK) {
        printf("Error processing %s: %d\n", argv[1], error);
        return 1;
    }
    printf("Found %d records\n", my_count);
    return 0;
}

Example: Convert a DTA to a tab-separated file.

#include "readstat.h"

int handle_metadata(readstat_metadata_t *metadata, void *ctx) {
    int *my_var_count = (int *)ctx;
    
    *my_var_count = readstat_get_var_count(metadata);

    return READSTAT_HANDLER_OK;
}

int handle_variable(int index, readstat_variable_t *variable, 
    const char *val_labels, void *ctx) {
    int *my_var_count = (int *)ctx;

    printf("%s", readstat_variable_get_name(variable));
    if (index == *my_var_count - 1) {
        printf("\n");
    } else {
        printf("\t");
    }

    return READSTAT_HANDLER_OK;
}

int handle_value(int obs_index, readstat_variable_t *variable, readstat_value_t value, void *ctx) {
    int *my_var_count = (int *)ctx;
    int var_index = readstat_variable_get_index(variable);
    readstat_type_t type = readstat_value_type(value);
    if (!readstat_value_is_system_missing(value)) {
        if (type == READSTAT_TYPE_STRING) {
            printf("%s", readstat_string_value(value));
        } else if (type == READSTAT_TYPE_INT8) {
            printf("%hhd", readstat_int8_value(value));
        } else if (type == READSTAT_TYPE_INT16) {
            printf("%hd", readstat_int16_value(value));
        } else if (type == READSTAT_TYPE_INT32) {
            printf("%d", readstat_int32_value(value));
        } else if (type == READSTAT_TYPE_FLOAT) {
            printf("%f", readstat_float_value(value));
        } else if (type == READSTAT_TYPE_DOUBLE) {
            printf("%lf", readstat_double_value(value));
        }
    }
    if (var_index == *my_var_count - 1) {
        printf("\n");
    } else {
        printf("\t");
    }

    return READSTAT_HANDLER_OK;
}

int main(int argc, char *argv[]) {
    if (argc != 2) {
        printf("Usage: %s <filename>\n", argv[0]);
        return 1;
    }
    int my_var_count = 0;
    readstat_error_t error = READSTAT_OK;
    readstat_parser_t *parser = readstat_parser_init();
    readstat_set_metadata_handler(parser, &handle_metadata);
    readstat_set_variable_handler(parser, &handle_variable);
    readstat_set_value_handler(parser, &handle_value);

    error = readstat_parse_dta(parser, argv[1], &my_var_count);

    readstat_parser_free(parser);

    if (error != READSTAT_OK) {
        printf("Error processing %s: %d\n", argv[1], error);
        return 1;
    }
    return 0;
}

ReadStat can write data sets to a number of file formats, and uses largely the same API for each of them. Files are written incrementally, with the header written first, followed by individual rows of data, and ending with some kind of trailer. (So the full data file never resides in memory.) Unlike like the callback-based API for reading files, the writer API consists of function that the developer must call in a particular order. The complete API can be found in readstat.h.

Basic usage:

#include "readstat.h"

/* A callback for writing bytes to your file descriptor of choice */
/* The ctx argument comes from the readstat_begin_writing_xxx function */
static ssize_t write_bytes(const void *data, size_t len, void *ctx) {
    int fd = *(int *)ctx;
    return write(fd, data, len);
}

int main(int argc, char *argv[]) {
    readstat_writer_t *writer = readstat_writer_init();
    readstat_set_data_writer(writer, &write_bytes);
    readstat_writer_set_file_label(writer, "My data set");

    int row_count = 1;

    readstat_variable_t *variable = readstat_add_variable(writer, "Var1", READSTAT_TYPE_DOUBLE, 0);
    readstat_variable_set_label(variable, "First variable");

    /* Call one of:
     *   readstat_begin_writing_dta
     *   readstat_begin_writing_por
     *   readstat_begin_writing_sas7bdat
     *   readstat_begin_writing_sav
     *   readstat_begin_writing_xport
     */

    int fd = open("something.dta", O_CREAT | O_WRONLY);
    readstat_begin_writing_dta(writer, &fd, row_count);

    int i;
    for (i=0; i<row_count; i++) {
        readstat_begin_row(writer);
        readstat_insert_double_value(writer, variable, 1.0 * i);
        readstat_end_row(writer);
    }

    readstat_end_writing(writer);
    readstat_writer_free(writer);
    close(fd);

    return 0;
}

To assist in fuzz testing, ReadStat ships with target files designed to work with libFuzzer. Clang 6 or later is required.

1. ./configure --enable-fuzz-testing turns on useful sanitizer and sanitizer-coverage flags
2. make will create a new binary called generate_corpus. Running this program will use the ReadStat test suite to create a corpus of test files in corpus/. There is a subdirectory for each sub-format (dta104, dta105, etc.). Currently a total of 468 files are created.
3. If fuzz-testing has been enabled, make will also create fourteen fuzzer targets, one for each of seven file formats, six for internally used grammars, and two fuzzers for testing the compression routines.
   • fuzz_format_dta
   • fuzz_format_por
   • fuzz_format_sas7bcat
   • fuzz_format_sas7bdat
   • fuzz_format_sav
   • fuzz_format_xport
   • fuzz_format_stata_dictionary
   • fuzz_grammar_dta_timestamp
   • fuzz_grammar_por_double
   • fuzz_grammar_sav_date
   • fuzz_grammar_sav_time
   • fuzz_grammar_spss_format
   • fuzz_grammar_xport_format
   • fuzz_compression_sas_rle
   • fuzz_compression_sav

For best results, each sub-directory of the corpus should be passed to the relevant fuzzer, e.g.:

• ./fuzz_format_dta corpus/dta104
• ./fuzz_format_dta corpus/dta110
• ...
• ./fuzz_format_sav corpus/sav
• ./fuzz_format_sav corpus/zsav
• ./fuzz_format_xport corpus/xpt5
• ./fuzz_format_xport corpus/xpt8

Finally, the compression fuzzers can be invoked without a corpus:

• ./fuzz_compression_sas_rle
• ./fuzz_compression_sav