Comments/PIs before doctype are lost

[Olli Pottonen]

>>> import lxml.etree
>>> doc = lxml.etree.fromstring('<?xml version="1.0"?><!--foo--><!DOCTYPE a><a/>').getroottree()
>>> etree.tostring(doc)

Expected result:
'<!--foo--><!DOCTYPE a>\n<a/>'
actual result:
'<!DOCTYPE a>\n<a/>'

Version info:
Python : sys.version_info(major=2, minor=7, micro=3, releaselevel='lxml.etree : (2, 3, 2, 0)
libxml used : (2, 8, 0)
libxml compiled : (2, 8, 0)
libxslt used : (1, 1, 26)
libxslt compiled : (1, 1, 26)

[Olli Pottonen (olli-pottonen)]

The comment is parsed correctly and doc.getroottree().getprevious() returns it as expected. The bug is in serialization. _writePrevSiblings() in serializer.pxi omits declaration (as it should) and everything before it (as it should not).

Because lxml handles the declaration itself, instead of relying on libxml2, this is difficult to get right.

[Olli Pottonen (olli-pottonen)]

I was wrong, it is not difficult to get right, just couple of additional lines of code.

[scoder (scoder)]

https://github.com/lxml/lxml/commit/ed1c31a2b9647b150445e00aabc729aef2c41f2a

[scoder (scoder)]

Fixed in lxml 3.5.0.