Output xsd:double

[dukesook]

I'm trying to write a literal xsd:double in here:

from rdflib import Graph, Literal, Namespace, URIRef, XSD
g = Graph()
example = Namespace("http://example.org/")
subject = URIRef("http://example/24")
double_literal = Literal("12345.678912345", datatype=XSD.double)
g.add((subject, example["foo"], double_literal))
print(g.serialize(format='turtle'))

Output

@prefix ns1: <http://example.org/> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

<http://example/24> ns1:foo 1.234568e+04 .

However, instead of "12345.678912345"^^xsd:double, I'm getting 1.234568e+04. I want to explicitly show the datatype and not lose any precision.

[ashleysommer]

Thanks for filing this issue @dukesook
I've looked into it, and what I found is actually alarming.

When serializing any literal to turtle representation (aka, N3 serialization), a method called _literal_n3 is called.
It looks like _literal_n3() can have an argument to serialize in "plain" mode, that will "shorten" the representation of Literals in a Turtle document. That includes dropping datatypes where not required, and in the case of floats and doubles, it shortens them to 6 decimal places of precision.

rdflib/rdflib/term.py

Lines 1528 to 1530 in 0b69f4f

	Only limited precision available for floats:
	>>> Literal(0.123456789)._literal_n3(use_plain=True)
	'1.234568e-01'

I don't know why anyone would want to do that deliberately write incorrect data to the output file.
But it looks like the RDFlib built-in turtle serializer will always use "plain" formatting mode for all literals written out:

rdflib/rdflib/plugins/serializers/turtle.py

Lines 360 to 364 in 0b69f4f

	if isinstance(node, Literal):
	return node._literal_n3(
	use_plain=True,
	qname_callback=lambda dt: self.getQName(dt, _GEN_QNAME_FOR_DT),
	)

I don't why it does this by default, or why its not configurable, but it seems like RDFLib has operated in this way since before the Turtle serializer was moved into the module in RDFLib codebase in RDFLib v3.0.0 in 2010.
0ffb306

[ashleysommer]

I found where the 6 decimal places comes from.
Its Python's built-in scientific-notation format for floats (and doubles)
https://docs.python.org/3/library/string.html#formatspec

When you use the "%e" formatting mark (or ":e" in f-strings), it converts the double to scientific notation, with "p" decimal points. Where if "p" is not given, it defaults to 6.

We can't simply change it to something higher like f"{float(value)}:.9e" because then a simpler double like 3.14 would serialize to "3.140000000e+01".

[ashleysommer]

Making a fix for this would be considered a breaking change, so this is another candidate for something to overhaul for the big RDFLib 8 release later this year.

[dukesook]

@ashleysommer
Thanks for looking into this issue, I look forward to it getting fixed.