Text node ending with non-breaking space incorrectly handled

[TimG1964]

XML v0.3.5

julia> xml1 = "<si><t>Ends with non-breaking space 0x00A0 </t></si>"
"<si><t>Ends with non-breaking space 0x00A0 </t></si>"

julia> using XML

julia> d=XML.parse(XML.LazyNode, xml1)
LazyNode (depth=0) Document

julia> n=next(d)
LazyNode (depth=1) Element <si>

julia> n=next(n)
LazyNode (depth=2) Element <t>

julia> n=next(n)
LazyNode (depth=3) Text "Ends with non-breaking space 0x00A0\xc2"

julia> 

The non-breaking space consists of two bytes: Char: ' ' → Code units: 2 → Bytes: UInt8[0xc2, 0xa0]
The second of these (0xa0) is recognised by isspace as a space and so the previous byte (0xc2) incorrectly remains in the text.

It is not clear to me if this is an issue with XML.jl or with Base.isspace, but it is easily fixable in XML.jl

[TimG1964]

A bit more research shows that non-breaking spaces are generally considered whitespace, so Base.isspace is correct:

Base.isspace(Char(0x00a0))
true

XML splits the two bytes, though, and this is the source of the issue.

However, the XML spec makes xml an exception in handling non-breaking spaces and says non-breaking spaces are considered to be meaningful content and should be retained as part of any text - regardless of the status of xml:space. This means that XML.jl probably shouldn't rely on Base.isspace anyway.

The XML spec seems to be relatively straightforward though. Here is a simple suggestion to replace Base.isspace:

#isspace(x::UInt8) = Base.isspace(Char(x))

# XML whitespace per XML 1.0/1.1 production S:
#   S ::= (#x20 | #x9 | #xD | #xA)+
@inline xml_isspace(b::UInt8)::Bool = (b == 0x20) | (b == 0x09) | (b == 0x0A) | (b == 0x0D)

(I've suggested changing the name to avoid possible piracy, since XML.jl doesn't own UInt8.)