diff --git a/parserInternals.c b/parserInternals.c
index af7ba45c..a805a13f 100644
--- a/parserInternals.c
+++ b/parserInternals.c
@@ -1340,7 +1340,16 @@ xmlInputSetEncodingHandler(xmlParserInputPtr input,
         input->consumed += processed;
         in->rawconsumed = processed;
 
-        nbchars = 4000 /* MINLEN */;
+        /*
+         * If we're push-parsing, we must convert the whole buffer.
+         *
+         * If we're pull-parsing, we could be parsing from a huge
+         * memory buffer which we don't want to convert completely.
+         */
+        if (input->flags & XML_INPUT_PROGRESSIVE)
+            nbchars = SIZE_MAX;
+        else
+            nbchars = 4000 /* MINLEN */;
         res = xmlCharEncInput(in, &nbchars);
         if (res < 0)
             code = in->error;
diff --git a/testparser.c b/testparser.c
index 6dfb7ea8..561f644f 100644
--- a/testparser.c
+++ b/testparser.c
@@ -340,6 +340,36 @@ testHugeEncodedChunk(void) {
     xmlFreeParserCtxt(ctxt);
     xmlFree(chunk);
 
+    /*
+     * Test the push parser with
+     *
+     * - a single call to xmlParseChunk,
+     * - a non-UTF8 encoding,
+     * - a chunk larger then MINLEN (4000 bytes).
+     *
+     * This verifies that the whole buffer is processed in the initial
+     * charset conversion.
+     */
+    buf = xmlBufferCreate();
+    xmlBufferCat(buf,
+            BAD_CAST "<?xml version='1.0' encoding='ISO-8859-1'?>\n");
+    xmlBufferCat(buf, BAD_CAST "<doc>");
+    /* 20,000 characters */
+    for (i = 0; i < 2000; i++)
+        xmlBufferCat(buf, BAD_CAST "0123456789");
+    xmlBufferCat(buf, BAD_CAST "</doc>");
+    chunk = xmlBufferDetach(buf);
+    xmlBufferFree(buf);
+
+    ctxt = xmlCreatePushParserCtxt(NULL, NULL, NULL, 0, NULL);
+
+    xmlParseChunk(ctxt, (char *) chunk, xmlStrlen(chunk), 1);
+
+    err = ctxt->wellFormed ? 0 : 1;
+    xmlFreeDoc(ctxt->myDoc);
+    xmlFreeParserCtxt(ctxt);
+    xmlFree(chunk);
+
     return err;
 }
 #endif