Redo path parsing for non-special URLs

[annevk]

As part of #148 and #185 we noticed that if we want to keep parsing non-special URLs in a way that preserves the ability for them to be base/relative URLs we need to make some changes, that would also bring them closer to the original RFCs.

I tried to do that as part of #185 but failed multiple times, so I'm going to make it a dedicated issue. What I think needs to happen is that we stop eating "/" early on. I think we need to drop the "path start state" as well.

What we want is that for special URLs we ensure that path has at least one item (potentially the empty string) and that for non-special URLs it can be empty if there is no path (e.g., as in foo://x#x having no path, but http://x#x becoming http://x/#x because special).

Currently "relative slash state" goes down "special authority ignore slashes state" even for non-special URLs. That means ///x against foo://y results in foo://x rather than foo:///x. That also needs to be fixed.

[annevk]

With these changes I get passing tests again (apart from five tests that require the opaque host changes to land):

diff --git a/lib/URL-impl.js b/lib/URL-impl.js
index 9e7d67c..0a394a3 100644
--- a/lib/URL-impl.js
+++ b/lib/URL-impl.js
@@ -136,6 +136,10 @@ exports.implementation = class URLImpl {
       return this._url.path[0];
     }
 
+    if (this._url.path.length === 0) {
+      return "";
+    }
+
     return "/" + this._url.path.join("/");
   }
 
diff --git a/src/url-state-machine.js b/src/url-state-machine.js
index 4534bd6..85bb11d 100644
--- a/src/url-state-machine.js
+++ b/src/url-state-machine.js
@@ -680,11 +680,13 @@ URLStateMachine.prototype["parse relative"] = function parseRelative(c) {
 };
 
 URLStateMachine.prototype["parse relative slash"] = function parseRelativeSlash(c) {
-  if (c === p("/") || (isSpecial(this.url) && c === p("\\"))) {
+  if (isSpecial(this.url) && (c === p("/") || c === p("\\"))) {
     if (c === p("\\")) {
       this.parseError = true;
     }
     this.state = "special authority ignore slashes";
+  } else if(c === p("/")) {
+    this.state = "authority";
   } else {
     this.url.username = this.base.username;
     this.url.password = this.base.password;
@@ -930,12 +932,27 @@ URLStateMachine.prototype["parse file host"] = function parseFileHost(c, cStr) {
 };
 
 URLStateMachine.prototype["parse path start"] = function parsePathStart(c) {
-  if (isSpecial(this.url) && c === p("\\")) {
-    this.parseError = true;
-  }
-  this.state = "path";
-  if (c !== p("/") && !(isSpecial(this.url) && c === p("\\"))) {
-    --this.pointer;
+  if (!isSpecial(this.url)) {
+    if (!this.stateOverride && c === p("?")) {
+      this.url.query = "";
+      this.state = "query";
+    } else if (!this.stateOverride && c === p("#")) {
+      this.url.fragment = "";
+      this.state = "fragment";
+    } else if (!isNaN(c)) {
+      if (c !== p("/")) {
+        --this.pointer;
+      }
+      this.state = "path";
+    }
+  } else {
+    if (c !== p("/") && c !== p("\\")) {
+      --this.pointer;
+    }
+    if(c === p("\\")) {
+      this.parseError = true;
+    }
+    this.state = "path";
   }
 
   return true;
@@ -1096,7 +1113,7 @@ function serializeURL(url, excludeFragment) {
 
   if (url.cannotBeABaseURL) {
     output += url.path[0];
-  } else {
+  } else if (url.path.length !== 0) {
     output += "/" + url.path.join("/");
   }

[annevk]

However, that probably means I should write more tests, since that seems like it would fail with state override into effect. Done and updated the patch above.