Fix processing instruction with attributes in formatting

[NikolasKomonen]

Fixes redhat-developer/vscode-xml#115

Also fixes an issue with the Prolog removing unknown attributes on format.

Signed-off-by: Nikolas [email protected]

[fbricon]

why do you need to treat xml-stylesheet differently than any other PI?

[NikolasKomonen]

@fbricon

Based off the wiki it looks like xml-stylesheet is a common processing instruction that is defined to have attributes. Other processing instructions have text content in them (<?mvn aaa bbb ccc ?>).

It helps in the parsing since we know how to parse after a certain PI name.

[fbricon]

In http://www.sagehill.net/docbookxsl/ProcessingInstructions.html (linked from the wiki), you can see

Processing instructions
A processing instruction is a means of passing hints to the XSLT processor about how to handle something. XML permits the placement of processing instructions just about anywhere in a document. A processing instruction (sometimes referred to as a PI) uses a special syntax, , to indicate that it is not an element, comment, or text. For example:

<?dbfo bgcolor="#EEEEEE" ?>
The first word in a PI is the name. In this example, the name of the processing instruction is dbfo and the content is bgcolor="#EEEEEE". This style of PI content could be called a “pseudo attribute”, because it provides a name-value pair similar to a real attribute (which only elements can have). You can put several such pseudo attributes in one PI, or you can put each of them in their own PI.

So xml-stylesheet should not be a special case

[NikolasKomonen]

@fbricon I know this is the case, but the regex in the xml specification considers the PI content as text.

What pr does is allow common PI declarations to be defined to be parsed with their attributes considered. If not then it defaults to saving all the content as a single string.

[fbricon]

can't you split content by whitespace and then check if you have key/value patterns in each group?

[NikolasKomonen]

@fbricon something like this wouldn't work in that case:

<?aaa b =   "123" c  ="234 ?>

To get around this it would require some backtracking in the parser, but we haven't done this yet for anything else.

[fbricon]

Should be the same behavior for all PIs

[NikolasKomonen]

@fbricon What if there exists a PI's that looks like

<?whatever foo   bar  = hello ?>
<?whatever this    should not be   formatted ?>
<?whatever This thing = no formatting ?>

at what point should it decide to format the content inside?

[fbricon]

any = sign you find indicates this is the middle of a key/value pair

[fbricon]

I still think all PIs should be treated like xml-stylesheet. If somehow we got a complaint, we can revisit, but that seems the most sensible, logical behavior.

[NikolasKomonen]

@fbricon Updated, all PI's are formatted the same (not touched)