Re: [9front] htmlfmt anchor corner cases

[umbraticus@prosimetrum.com]

> i do not like this part so much.
> 
> +		if(href!=nil && href[0]=='/'
> +		&& (base = strchr(result, ':')) != nil
> +		&& (base = strchr(base+3, '/')) != nil)
> +			*base = '\0';
> 
> the issue is htmlfmt's code to combine relative
> urls is just wrong. handling urls can be hard.
> 
> but adding hacks like these does not solve the
> problem sufficiently. theres code in webfs that
> might be of help.
> 
> maybe we should make a version for libhtml,
> which htmlfmt uses.

Yes, each program that uses libhtml does its own thing:

/sys/src/cmd/htmlfmt/html.c:/^fullurl
/sys/src/cmd/abaco/urls.c:/^urlcombine	quote: /* this is a HACK */
/sys/src/cmd/mothra/url.c:/^fileget
/sys/src/cmd/mothra/url.c:/^webclone ← makes use of webfs

Below is a patch that makes urls absolute when possible during parsing.
I'm not sure if the comment // FOR NOW: leave the url relative.
indicates that this was intended all along...

This would obviate the functions in htmlfmt and abaco but mothra
doesn't even use parsehtml so...  is it even worth it?  I'll have
another go at tidying up htmlfmt and send a separate email.

umbraticus

diff -r f4a5c13bcd43 sys/src/libhtml/build.c
--- a/sys/src/libhtml/build.c	Mon Dec 28 12:24:47 2020 +0100
+++ b/sys/src/libhtml/build.c	Wed Dec 30 16:12:33 2020 +1300
@@ -309,7 +309,6 @@
 static void			pushfontstyle(Pstate* ps, int sty);
 static void			pushjust(Pstate* ps, int j);
 static Item*		textit(Pstate* ps, Rune* s);
-static Rune*		removeallwhite(Rune* s);
 static void			resetdocinfo(Docinfo* d);
 static void			setcurfont(Pstate* ps);
 static void			setcurjust(Pstate* ps);
@@ -425,6 +424,40 @@
 
 static Item *getitems(ItemSource* is, uchar* data, int datalen);
 
+// Return malloced url, given (possibly empty) path and base.
+// A relative path and absolute base are combined; otherwise, path is returned as is.
+// If path is nil and base absolute, base up to final slash is returned.
+// URL strings are not validated any further than checking for proto://
+Rune*
+_fullurl(Rune *path, Rune *base)
+{
+	Rune *r;
+
+	if(path != nil){
+		for(r = path; isalpha(*r); r++)
+			;
+		if(r > path && *r++ == ':' && *r++ == '/' && *r == '/')
+			return _Strdup(path);	/* path is already absolute */
+	}
+	if(base == nil)
+		return _Strdup(path);
+	for(r = base; isalpha(*r); r++)
+		;
+	if(r == base || *r++ != ':' || *r++ != '/' || *r++ != '/')
+		return _Strdup(path);	/* bad base url proto */
+	while(isalnum(*r) || *r == '_' || *r == '@' || *r == '-' || *r == ':' || *r == '.')
+		r++;
+	if(r[-1] == '/' || *r && *r != '/')
+		return _Strdup(path);	/* bad base url hostname */
+	if(*r == '/' && (path == nil || *path != '/'))
+		r = runestrrchr(r, '/');	/* find final slash if path is not rooted */
+	if(path == nil)
+		return runesmprint("%.*S/", (int)(r - base), base);
+	if(*path == '/')
+		path++;
+	return runesmprint("%.*S/%S", (int)(r - base), base, path);
+}
+
 // Parse an html document and create a list of layout items.
 // Allocate and return document info in *pdi.
 // When caller is done with the items, it should call
@@ -439,7 +472,7 @@
 
 	di = newdocinfo();
 	di->src = _Strdup(pagesrc);
-	di->base = _Strdup(pagesrc);
+	di->base = _fullurl(nil, pagesrc);
 	di->mediatype = mtype;
 	di->chset = chset;
 	*pdi = di;
@@ -2923,55 +2956,18 @@
 	return ans;
 }
 
-// Attribute value when value is a URL, possibly relative to base.
-// FOR NOW: leave the url relative.
+// Attribute value when value is a URL.
+// Relative URLs are converted to absolute if a suitable base is given.
 // Caller must free the result (eventually).
 static Rune*
 aurlval(Token* tok, int attid, Rune* dflt, Rune* base)
 {
 	Rune*	ans;
-	Rune*	url;
-
-	USED(base);
-	ans = nil;
-	if(_tokaval(tok, attid, &url, 0) && url != nil)
-		ans = removeallwhite(url);
+
+	_tokaval(tok, attid, &ans, 0);
 	if(ans == nil)
-		ans = _Strdup(dflt);
-	return ans;
-}
-
-// Return copy of s but with all whitespace (even internal) removed.
-// This fixes some buggy URL specification strings.
-static Rune*
-removeallwhite(Rune* s)
-{
-	int	j;
-	int	n;
-	int	i;
-	int	c;
-	Rune*	ans;
-
-	j = 0;
-	n = _Strlen(s);
-	for(i = 0; i < n; i++) {
-		c = s[i];
-		if(c >= 256 || !isspace(c))
-			j++;
-	}
-	if(j < n) {
-		ans = _newstr(j);
-		j = 0;
-		for(i = 0; i < n; i++) {
-			c = s[i];
-			if(c >= 256 || !isspace(c))
-				ans[j++] = c;
-		}
-		ans[j] = 0;
-	}
-	else
-		ans = _Strdup(s);
-	return ans;
+		return _Strdup(dflt);
+	return _fullurl(ans, base);
 }
 
 // Attribute value when mere presence of attr implies value of 1,