ISO-8859-1 is used as default encoding

[Rogach]

ISO-8859-1 (also known as latin1) is used as default encoding of XML.save:
https://github.com/scala/scala-xml/blob/master/src/main/scala/scala/xml/XML.scala#L67

This results in errors if document contains non-latin1 characters. Why this encoding is used by default instead of UTF-8?

[ashawley]

Why this encoding is used by default instead of UTF-8?

Presumably for backwards compatability.

In 2003, ISO-8859-1 became the default for save in 5d19bda.

Changing the default to UTF-8 would be a good improvement to make in a major release.

Would it help to add a mention of ISO-8859-1 to the documentation, with an example of using UTF-8?

    scala.xml.XML.save("foo.xml", <foo/>, "UTF-8")

[ashawley]

That ISO-8859-1 was the default, was mentioned in Scala 2.6.1:

http://www.scala-lang.org/api/2.6.1/scala/xml/XML$object.html

Today, ISO-8859-1 is not mentioned:

http://scala-lang.org/files/archive/api/2.12.0/scala-xml/scala/xml/XML$.html

It was removed in 31b4fe1.

[Rogach]

I understand the value of backwards compatibility, but sadly the next major release would probably never happen.

Yes, adding a mention to documentation would definitely help! Actually, if value was displayed in actual function signature (enc: String = "ISO-8859-1") it would be immediately obvious, but since it is factored out into value scaladoc only displays enc: String = encoding.

Should I make a pull request with this change?

P.S. "Another big XML commit" - and here I am, teaching people about value of good commit messages =)

[SethTisue]

I'm not sure I agree that an actual fix would need to wait until the next major version

[Rogach]

How else would it be possible? It would probably break code that depends on latin1 characters outside of ASCII range.

[SethTisue]

I guess it depends how much meaning you attach to the version number. I'm not a semver purist, but some are. I would argue that user code that didn't specify an encoding had a latent bug in it and we should feel free to break that code, with a prominent explanation in the release note.

Basically I think we should ship a fix sooner rather than later. I don't have a strong opinion about the version number.

[Rogach]

I, for one, am very much for it - my code would not break :)

[ashawley]

The value of encoding in save is used in the XML declaration:

<?xml version='1.0' encoding='ISO-8859-1'?>

However, the default is not to put in an XML declaration. Only if someone did one of these two would the explicit mention of ISO-8859-1 be included:

scala.xml.XML.save("foo.xml", <foo/>, "ISO-8859-1", true)
scala.xml.XML.save("foo.xml", <foo/>, xmlDecl = true)

The default also writes bytes in ISO-8859-1. As rogach alluded, ASCII is forward compatible with UTF-8, but ISO-8859-1 is not.

According to the XML spec, "parsed entities which are stored in an encoding other than UTF-8 or UTF-16 must begin with a text declaration". So writing something other than ASCII, UTF-8 or UTF-16 without an appropriate declaration, as this library does, is not compliant.

Here's proof from our friend, the Xerces parsing library, that chokes on a simple example:

scala> val latin1 = ((32 to 126).toStream ++ (160 to 255)).map(_.toChar)
latin1: scala.collection.immutable.Stream[Char] = Stream( , ?)

scala> scala.xml.XML.save("foo.xml", <x>{ latin1.mkString }</x>)

scala> scala.xml.XML.loadFile("foo.xml")
com.sun.org.apache.xerces.internal.impl.io.MalformedByteSequenceException: Invalid byte 1 of 1-byte UTF-8 sequence.
...

Adding a declaration, with xmlDecl set to true, allows it to be parsed:

scala> scala.xml.XML.save("foo.xml", <x>{ latin1.mkString }</x>, xmlDecl = true)

scala> scala.xml.XML.loadFile("foo.xml")
res3: scala.xml.Elem = <x> !&quot;#$%&amp;'()*+,-./0123456789:;&lt;=&gt;?
@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz
{|}~ ¡¢£¤¥¦§¨©ª«¬­®¯°±²³´µ¶·¸¹º»¼½¾¿ÀÁÂÃÄÅÆÇÈÉÊËÌÍÎÏÐÑÒÓÔÕÖ×
ØÙÚÛÜÝÞßàáâãäåæçèéêëìíîïðñòóôõö÷øùúûüýþÿ</x>

The proposal to make UTF-8 default, would make save compliant, and fix the bug:

scala> scala.xml.XML.save("foo.xml", <x>{ latin1.mkString }</x>, "UTF-8")

res3: scala.xml.Elem = <x> !&quot;#$%&amp;'()*+,-./0123456789:;&lt;=&gt;?
@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz
{|}~ ¡¢£¤¥¦§¨©ª«¬­®¯°±²³´µ¶·¸¹º»¼½¾¿ÀÁÂÃÄÅÆÇÈÉÊËÌÍÎÏÐÑÒÓÔÕÖ×
ØÙÚÛÜÝÞßàáâãäåæçèéêëìíîïðñòóôõö÷øùúûüýþÿ</x>

It is a "latent bug", as Seth suggests. For users of save not specifying an encoding, fixing the default to be UTF-8 could introduce bugs in their code, but that depends on

1. if they're using the high-order bits of ISO-8859-1, and
2. what is consuming the invalid XML they produce.

There's a chance that something in (2) could behave in a more permissive or dumber way than Xerces did above.

[Rogach]

@ashawley - awesome summary!

I think that most other libs will expect utf-8 as default by now, so keeping latin1 as default encoding only creates more latent bugs in the wild.

Question to developers: should I make a pull request with the change or somebody else will do it?

[ashawley]

Yeah, seems it's worth a PR that fixes the issue. Thank you

[SethTisue]

Should it default to java.nio.charset.Charset.defaultCharset, or to some fixed default such as UTF-8?

[Rogach]

I would prefer fixed default - it is saner and less prone to surprises. And I believe everybody should use UTF-8 as default charset anyway.

[Rogach]

Here's the PR: #122