made more robust html discovery by using htmltidy

 now when parsing document that we expect might be html we also save first 16K
 of the document to the buffer and if the parser choked we run the saved data
 through htmltidy and feed the output to the parser again.

Signed-off-by: Michael Krelin <hacker@klever.net>

1 files changed, 62 insertions, 9 deletions

diff --git a/lib/discovery.cc b/lib/discovery.cc
index 8729cfb..a308b56 100644
--- a/lib/discovery.cc
+++ b/lib/discovery.cc
@@ -1,33 +1,35 @@
 #include <list>
 #include <opkele/curl.h>
 #include <opkele/expat.h>
 #include <opkele/uris.h>
 #include <opkele/discovery.h>
 #include <opkele/exception.h>
 #include <opkele/util.h>
+#include <opkele/tidy.h>
 #include <opkele/debug.h>
 
 #include "config.h"
 
 #define XRDS_HEADER "X-XRDS-Location"
 #define CT_HEADER "Content-Type"
 
 namespace opkele {
     using std::list;
     using xrd::XRD_t;
     using xrd::service_t;
 
     static const char *whitespace = " \t\r\n";
     static const char *i_leaders = "=@+$!(";
+    static const size_t max_html = 16384;
 
     static inline bool is_qelement(const XML_Char *n,const char *qen) {
 	return !strcasecmp(n,qen);
     }
     static inline bool is_element(const XML_Char *n,const char *en) {
 	if(!strcasecmp(n,en)) return true;
 	int nl = strlen(n), enl = strlen(en);
 	if( (nl>=(enl+1)) && n[nl-enl-1]=='\t'
 		&& !strcasecmp(&n[nl-enl],en) )
 	    return true;
 	return false;
     }
@@ -52,24 +54,26 @@ namespace opkele {
 
 	    string xrds_location;
 	    string http_content_type;
 	    service_t html_openid1;
 	    service_t html_openid2;
 	    string cdata_buf;
 	    long status_code;
 	    string status_string;
 
 	    typedef list<string> pt_stack_t;
 	    pt_stack_t pt_stack;
 	    int skipping;
+	    bool parser_choked;
+	    string save_html;
 
 	    XRD_t *xrd;
 	    service_t *xrd_service;
 	    string* cdata;
 
 	    idigger_t()
 		: util::curl_t(easy_init()),
 		util::expat_t(0),
 		xri_proxy(XRI_PROXY_URL) {
 		    CURLcode r;
 		    (r=misc_sets())
 			|| (r=set_write())
@@ -131,63 +135,112 @@ namespace opkele {
 			discover_at(result,xrds_location,xmode_xrd);
 			if(result.xrd.empty())
 			    html2xrd(result.xrd);
 		    }
 		}
 	    }
 
 	    void discover_at(idiscovery_t& result,const string& url,int xm) {
 		CURLcode r = easy_setopt(CURLOPT_URL,url.c_str());
 		if(r)
 		    throw exception_curl(OPKELE_CP_ "failed to set culry urlie",r);
 
-		(*(expat_t*)this) = parser_create_ns();
-		set_user_data(); set_element_handler();
-		set_character_data_handler();
-
 		http_content_type.clear();
 		xmode = xm;
+		prepare_to_parse();
 		if(xmode&xmode_html) {
 		    xrds_location.clear();
-		    html_openid1.clear(); html_openid2.clear();
+		    save_html.clear();
+		    save_html.reserve(max_html);
 		}
 		xrd = &result.xrd;
-		cdata = 0; xrd_service = 0; skipping = 0;
-		status_code = 100; status_string.clear();
 
 		r = easy_perform();
 		if(r && r!=CURLE_WRITE_ERROR)
 		    throw exception_curl(OPKELE_CP_ "failed to perform curly request",r);
 
+		if(!parser_choked) {
 		parse(0,0,true);
+		}else{
+		    /* TODO: do not bother if we've seen xml */
+		    try {
+			util::tidy_doc_t td = util::tidy_doc_t::create();
+			if(!td)
+			    throw exception_tidy(OPKELE_CP_ "failed to create htmltidy document");
+#ifndef NDEBUG
+			td.opt_set(TidyQuiet,false);
+			td.opt_set(TidyShowWarnings,false);
+			td.opt_set(TidyForceOutput,true);
+			td.opt_set(TidyXhtmlOut,true);
+			td.opt_set(TidyDoctypeMode,TidyDoctypeOmit);
+			td.opt_set(TidyMark,false);
+#endif /* NDEBUG */
+			if(td.parse_string(save_html)<=0)
+			    throw exception_tidy(OPKELE_CP_ "tidy failed to parse document");
+			if(td.clean_and_repair()<=0)
+			    throw exception_tidy(OPKELE_CP_ "tidy failed to clean and repair");
+			util::tidy_buf_t tide;
+			if(td.save_buffer(tide)<=0)
+			    throw exception_tidy(OPKELE_CP_ "tidy failed to save buffer");
+			prepare_to_parse();
+			parse(tide.c_str(),tide.size(),true);
+		    }catch(exception_tidy& et) { }
+		}
+		save_html.clear();
+	    }
+
+	    void prepare_to_parse() {
+		(*(expat_t*)this) = parser_create_ns();
+		set_user_data(); set_element_handler();
+		set_character_data_handler();
+
+		if(xmode&xmode_html) {
+		    html_openid1.clear(); html_openid2.clear();
+		    parser_choked = false;
+		}
+
+		cdata = 0; xrd_service = 0; skipping = 0;
+		status_code = 100; status_string.clear();
 	    }
 
 	    void html2xrd(XRD_t& x) {
 		if(!html_openid1.uris.empty()) {
 		    html_openid1.types.insert(STURI_OPENID11);
 		    x.services.add(-1,html_openid1);
 		}
 		if(!html_openid2.uris.empty()) {
 		    html_openid2.types.insert(STURI_OPENID20);
 		    x.services.add(-1,html_openid2);
 		}
 	    }
 
 	    size_t write(void *p,size_t s,size_t nm) {
-		if(skipping<0) return 0;
 		/* TODO: limit total size */
 		size_t bytes = s*nm;
-		bool rp = parse((const char *)p,bytes,false);
+		const char *inbuf = (const char*)p;
+		if(xmode&xmode_html) {
+		    size_t mbts = save_html.capacity()-save_html.size();
+		    size_t bts = 0;
+		    if(mbts>0) {
+			bts = (bytes>mbts)?mbts:bytes;
+			save_html.append(inbuf,bts);
+		    }
+		    if(skipping<0) return bts;
+		}
+		if(skipping<0) return 0;
+		bool rp = parse(inbuf,bytes,false);
 		if(!rp) {
+		    parser_choked = true;
 		    skipping = -1;
+		    if(!(xmode&xmode_html))
 		    bytes = 0;
 		}
 		return bytes;
 	    }
 	    size_t header(void *p,size_t s,size_t nm) {
 		size_t bytes = s*nm;
 		const char *h = (const char*)p;
 		const char *colon = (const char*)memchr(p,':',bytes);
 		const char *space = (const char*)memchr(p,' ',bytes);
 		if(space && ( (!colon) || space<colon ) ) {
 		    xrds_location.clear(); http_content_type.clear();
 		}else if(colon) {