Abstract

The formats CSV [[RFC4180]] (comma separated values) and TSV [[IANA-TSV]] (tab separated values) provide simple, easy to process formats for the transmission of tabular data. They are supported as input datat formats by many tools, particularly spreadsheets. This document describes their use for expressing SPARQL query results from SELECT queries.

This specification is published by the RDF Star Working Group as part of the update of specifications for format and errata.

Introduction

This document describes CSV and TSV formats for expressing the results of a SPARQL SELECT query. They provide lowest common denominator formats between systems using different implementation technologies.

Other formats for expression of SPARQL results are the SPARQL Results XML Format [[SPARQL12-RESULTS-XML]] and the SPARQL Results JSON Format [[SPARQL12-RESULTS-JSON]]. Each format is useful in different application scenarios.

The SPARQL Results CSV Format is a lossy encoding of a table of results. It does not encode all the details of each RDF term in the results; instead, it just gives a string without indicating the type of the term (IRI, Literal, Literal with datatype, Literal with language, or blank node). This makes it simple to consume data, such as text and numbers, in applications that don't need to understand the details of RDF. In some applications, guesses as to which elements are hyperlinks are made pragmatically, for example, guessing that strings starting "http://" are links.

The SPARQL Results TSV Format does encode the details of RDF terms in the results table, by using the syntax that SPARQL [[SPARQL12-QUERY]] and Turtle [[RDF12-TURTLE]] use. An application receiving a TSV-encoded result set can split each line into elements of the result row, and extract all the details of the RDF terms it wishes to process by simple string processing, without a complete XML or JSON parser as may by required by the more complex SPARQL result formats.

When this document uses the words must, must not, should, should not, may and recommended, they must be interpreted as described in RFC 2119 [[RFC2119]].

Example

The following artificial example is used to illustrate the features of serializing results in each format.

x literal Comment (not part of the table)
<http://example/x> String An IRI and a string consisting of characters S-t-r-i-n-g
<http://example/x> String-with-dquote" String with a double quote in it.
_:b0 Blank node Blank node
Missing 'x' No RDF term for the x column
This row has no terms in it.
<http://example/x> No term in the literal column.
_:b1 "String-with-lang"@en An RDF literal with a language tag
_:b1 "String-with-lang-dir"@en--ltr An RDF literal with a directional language tag
_:b1 123 An RDF literal, datatype xsd:integer, and lexical form 123.

The following artificial example is used to illustrate the features of serializing results containing triple terms.

x triple Comment (not part of the table)
"Alice" <<( <http://example/alice> <http://example/knows> <http://example/bob> )>> A plain string and triple term with subject IRI http://example/alice, predicate IRI http://example/knows, and object IRI http://example/bob.
"Bob" <<( <http://example/bob> <http://example/knows> <http://example/alice> )>> A plain string and triple term with subject IRI http://example/bob, predicate IRI http://example/knows, and object IRI http://example/alice.
"Carol" <<( <http://example/carol> <http://example/says> "Hello world, my name is \"Alice\"." )>> A plain string and triple term with subject IRI http://example/carol, predicate IRI http://example/says, and object literal Hello world, my name is "Alice"..

Transmission issues using CSV and TSV Formats

The SPARQL result formats described here conform to the formal specifications of the relevant formats, Comma Separated values (CSV) [[RFC4180]] and Tab Separated Value (TSV) [[IANA-TSV]].

Systems providing these formats should note that the content types are text/csv for CSV and text/tab-separated-values for TSV. Being text/*, the default character set is US-ASCII. The charset parameter should be used in conjunction with SPARQL Results; UTF-8 is recommended; giving us text/csv; charset=utf-8 and text/tab-separated-values; charset=utf-8.

The end-of-line in CSV is CRLF, i.e., Unicode codepoints 13 (U+000D) and 10 (U+000A).

The end-of-line in TSV is EOL, i.e., Unicode codepoint 10 (U+000A).

Applications reading these formats are advised to cope with both CRLF and LF as end of line markers and not rely on conformance to the formal specifications.

CSV — Comma Separated values

In the SPARQL Results CSV Format, the results table is serialized as one line listing the variables in the results, using the CSV header line, followed by one line for each query solution. (Note: a line may end up split by newlines in the data). Values in the results are:

Serializing the Results Table

The first line of a SPARQL Results CSV Format response is the header line, giving the names of the variables used in the result set. The header line consists of the variable names, without leading question marks ?, separated by commas.

While the text/csv format does not require a header row, the SPARQL Results CSV Format must use a header row. If the content type parameter header is used, it must be header=present.

The remaining rows are the values of the results, with each binding determined by the position in the row, corresponding to the entry in the header line.

If a variable is not bound, an empty field is used (e.g. ,,). Each row must have the same number of fields, with each field corresponding to a binding to the variable in the header line in the same field position.

Serializing RDF Terms

The entry in each field is the string corresponding to the RDF term value. (cf. SPARQL STR()) without syntax to denote what kind of term it is. The encoding quoting rules of CSV format must be used.

Blank nodes use the _:label form from Turtle and SPARQL. Use of the same label indicates the same blank node within the result set but has no significance outside the result set.

Triple terms are enclosed in <<( … )>>, as <<( subject predicate object )>>, where the subject, predicate, and object terms are recursively serialized. The serializations of subject, predicate, and object must be separated by a single space character (SPACE, code point 32, U+0020). The single space character preceding subject, and the single space character following object are optional, and are mainly valuable for human readability. If the object is a literal, then it must be encapsulated within a pair of quotation marks "".

Fields containing any of " (QUOTATION MARK, code point 34, U+0022 in Unicode [[UNICODE]]), , (COMMA, code point 44, U+002C), LF (code point 10, U+000A), or CR (code point 13, U+000D) must be quoted using the quoting mechanism of RFC4180 [[RFC4180]]. Fields are delimited by a pair of quotation marks " (code point U+0022). Within quoted strings, all characters except ", including new line characters, have their exact meaning — newlines do not end a CSV record. Inline " is written using a pair of quotation marks "". This quoting mechanism is applied recursively for terms in triple terms.

Since literals in the object position of triple terms are encapsulated within a pair of quotation marks ", and the full triple term is always encapsulated within quotation marks as well, this requires the subsequent escaping of the "inner" quotation marks encapsulating the literal. Such cases will result in triple terms in the form of "<<( subject predicate ""object-literal"" )>>".

The standard CSV format does not distinguish between missing values and empty strings. The SPARQL 1.2 Results CSV Format uses the same representation for unbound variables as for variables bound to an empty string literal. The other SPARQL Result formats (based on JSON, TSV, or XML) can be used if this distinction is required.

Example of CSV-Serialized Results

x,literal
http://example/x,String
http://example/x,"String-with-dquote"""
_:b0,Blank node
,Missing 'x'
,
http://example/x,
_:b1,String-with-lang
_:b1,String-with-lang-dir
_:b1,123

Example of CSV-Serialized Results with Triple Terms

x,triple
"Alice",<<( http://example/alice http://example/knows http://example/bob )>>
"Bob",<<( http://example/bob http://example/knows http://example/alice )>>
"Carol","<<( http://example/carol http://example/says ""Hello world, my name is """"Alice""""."" )>>"

TSV — Tab Separated values

In the SPARQL Results TSV Format, the results table is serialized as one line listing the variables in the results, followed by one line for each query solution. All RDF terms used in the format are encoded in the format specified by Turtle [RDF12-TURTLE] except that the triple term forms for the lexical part of literals must not be used. These forms would allow raw newlines and tabs that are part of the TSV format. A TSV format SPARQL result set must use the single quoted literal forms, together with any necessary escapes such as \t, \n, and \r.

Serializing the Results Table

The results table is serialized as one line listing the variables in the results, followed by one line for each query solution. This first line is required by the TSV format [[IANA-TSV]], unlike CSV, where it is optional.

Variables are serialized in SPARQL syntax, using question mark ? character followed by the variable name.

Each row of the result set is serialized by sequence of RDF terms in SPARQL syntax, separated by a tab (Horizontal Tab, Unicode code point U+0009) character.

If a variable is not bound in a row, an empty field is used. Each row must have the same number of fields, corresponding to the variables listed in the first row.

Serializing RDF Terms

The SPARQL Results TSV Format serializes RDF terms in the results table by using the syntax that SPARQL [[SPARQL12-QUERY]] and Turtle [[RDF12-TURTLE]] use.

IRIs are enclosed in <...>, literals are enclosed with double quotes "..." or single quotes ' ...' with optional @lang or ^^ for datatype. IRIs are written enclosed in <...>. They must conform to the IRI rule of [[[RFC3987]]]. Such IRIs include the IRI scheme and must not be Relative Reference. This includes IRIs used as datatypes.

Literals are written with the lexical form in quotes. Tab, newline, and carriage return characters (Unicode code points U+0009 (tab), U+0010 (line feed) and U+0013 (carriage return)) are encoded in strings as \t, \n and \r respectively. The long string forms using triple quotes — """ or '''must not be used.

The abbreviated forms for numbers (XSD integers, decimals, and doubles) should be used.

Blank nodes use the _:label form from Turtle and SPARQL. Use of the same label indicates the same blank node within the result set, but has no significance outside the result set.

Triple terms are enclosed in <<( … )>>, as in <<(subject predicate object)>>, where the subject, predicate, and object terms are recursively serialized. The serializations of subject, predicate, and object must be separated by a single space character (SPACE, code point 32, U+0020). The single space character preceding subject, and the single space character following object are optional, and are mainly valuable for human readability.

Example of TSV-Serialized Results

Writing <TAB> for a raw tab character (Unicode code point U+0009):

?x<TAB>?literal
<http://example/x><TAB>"String"
<http://example/x><TAB>"String-with-dquote\"" 


_:blank0<TAB>"Blank node"
<TAB>"Missing 'x'"
<TAB>
<http://example/x><TAB>
_:blank1<TAB>"String-with-lang"@en
_:blank1<TAB>"String-with-lang-dir"@en--ltr
_:blank1<TAB>123

Example of TSV-Serialized Results with Triple Terms

Writing <TAB> for a raw tab character (Unicode code point U+0009):

?x<TAB>?triple
"Alice"<TAB><<( <http://example/alice> <http://example/knows> <http://example/bob> )>>
"Bob"<TAB><<( <http://example/bob> <http://example/knows> <http://example/alice> )>>
"Carol"<TAB><<( <http://example/carol> <http://example/says> "Hello world, my name is \"Alice\". )>>

Conformance

Changes between SPARQL 1.1 Query Results CSV and TSV Formats and SPARQL 1.2 Query Results CSV and TSV Formats

Privacy Considerations

TODO

Security Considerations

TODO

Internationalization Considerations

TODO