Re: Strange output of XML attribute values

st 16. 9. 2020 v 14:11 odesílatel Andrew Marynchuk (Андрей Маринчук) <radist.nt@gmail.com> napsal:

This problem is quite old, but it leads to the inability to use XML generation functions in PostgreSQL database for some cases, or at least requires to perform subsequent parsing and regenerating XML by an external utility. It reproduces in PostgreSQL 12.4, compiled by Visual C++ build 1914, 64-bit (windows 10), but I've seen the same problem in 9.6 build from CentOS yum package.

How to reproduce:

Just execute the query (actually the xmlelement call is enough to reproduce the proble):

select xmlserialize(document xmlroot(xmlelement(name "ЭлементВКириллице", xmlattributes('ЗначениеВКириллице' as "АтрибутВКириллице"), 'ТекстВКириллице'), version '1.0', standalone yes) as text);

Expected result:

<?xml version="1.0" standalone="yes"?><ЭлементВКириллице АтрибутВКириллице="ЗначениеВКириллице">ТекстВКириллице</ЭлементВКириллице>

Actual result:

<?xml version="1.0" standalone="yes"?><ЭлементВКириллице АтрибутВКириллице="&#x417;&#x43D;&#x430;&#x447;&#x435;&#x43D;&#x438;&#x435;&#x412;&#x41A;&#x438;&#x440;&#x438;&#x43B;&#x43B;&#x438;&#x446;&#x435;">ТекстВКириллице</ЭлементВКириллице>

This example uses cyrillic letters, but it could be any non-ASCII character.

According to the discussion, this problem arises because PostgreSQL does not provides libxml2 an information of document encoding due to the lack of xmlTextWriterStartDocument call, so libxml2 has no idea that encoding is UTF-8 and non-ASCII characters could be written without converting to &#x...;-sequences.

I don't think it is true. The url encoding is done only in attributes, and only when the output encoding will be utf8. When you try to use 8bit encoding with Azbuka support, it will be ok.

 

In the modern world, UTF-8 encoding is used everywhere and such unnecessary character converting looks strange. Current workaround is passing generated content to the pl/python function which parses and writes back the xml (xml.dom.minidom.parseString(...).toxml()).

If I remember some discussion about this topic, the problem is XML standard, that requires url encoding in attribute values.

I reported this issue 10 (maybe 15) years ago to libxml2 developers, and It was rejected. Maybe libxml2 supports too old XML standards. I don't know - this library is years in frozen state, but there is no replacement.

It is a libxml2 problem - , and there it should be reported and fixed. It is not possible to fix this issue on Postgres' side.

Regards

Pavel