Fix a regression in xmlGetDocCompressMode()

The switch to xzlib had for consequence that the compression
level of the input was not gathered anymore in ctxt->input->buf,
then the parser compression flags was left to -1 and propagated
to the resulting document.
Fix the I/O layer to get compression detection in xzlib,
then carry it in the input buffer and the resulting document

  This should fix
    https://lsbbugs.linuxfoundation.org/show_bug.cgi?id=3456

1 files changed, 4 insertions, 0 deletions

diff --git a/parser.c b/parser.c
index 1d478c3d..4a442bb9 100644
--- a/parser.c
+++ b/parser.c
@@ -10681,6 +10681,10 @@ xmlParseDocument(xmlParserCtxtPtr ctxt) {
         ctxt->sax->startDocument(ctxt->userData);
     if (ctxt->instate == XML_PARSER_EOF)
 	return(-1);
+    if ((ctxt->myDoc != NULL) && (ctxt->input != NULL) &&
+        (ctxt->input->buf != NULL) && (ctxt->input->buf->compressed >= 0)) {
+	ctxt->myDoc->compression = ctxt->input->buf->compressed;
+    }
 
     /*
      * The Misc part of the Prolog