Please describe the problem:
It appears that libxslt1.1 pays attention to the charset declaration in the Content-Type HTTP header when retrieving XML files with MIME types of application/xml or text/xml via the document() function.  If a misconfigured web server sends "Content-Type: text/xml; charset=iso-8859-15" but the XML file itself has no encoding declaration in the XML prolog (and is thus to be taken as UTF-8), libxslt treats the incoming file as ISO-8859-15 and so mangles byte sequences that express e.g. many common vowels with diacritics. libxslt does not exhibit the behavior when the MIME type is 'text/html'. Saxon 6.5.5 does not exhibit the same behavior with any MIME type/charset combination.

I am attaching a test stylesheet that takes itself as input, and retrieves a simple file in UTF-8 and Latin-9 encodings from a webserver, and outputs the results with MIME types and charsets noted.

Steps to reproduce:
1. Use the attached XSLT to transform itself, e.g. with 'xsltproc test.xsl test.xsl', and observe the output.


Actual results:
When the server gives the MIME type as 'application/xml' or 'text/xml' and the encoding as 'ISO-8859-15', the conversion from the ISO encoding to UTF-8 is applied to the UTF-8 document, resulting in mangled bytes in the 'text' element where the encodings differ.

Expected results:
I would expect the incoming files to be treated as UTF-8 always.

Does this happen every time?
Yes.

Other information: