Use getPath instead of getRawPath to prevent doulbe encoding of the URI

[jensmatw]

Use getPath instead of getRawPath to prevent doulbe encoding of the URI to fix issue #3482.

[rworsnop]

See #3437

[spencergibb]

please add a test

[rworsnop]

I'm not sure this is going to work.
Consider the url "http://example.com/foo–bar foo" (that's an en dash, not a dash):

 String url = "http://example.com/foo%E2%80%93bar%20foo";
 URI uri = URI.create(url);
 String path = uri.getPath();
 URI uri2 = UriComponentsBuilder.fromUri(uri).replacePath(path).build().toUri(); 

uri2 is "http://example.com/foo–bar%20foo", so the space got re-encoded properly, but the en dash did not.

[jensmatw]

I'm not sure this is going to work. Consider the url "http://example.com/foo–bar foo" (that's an en dash, not a dash):

You are right, it only solves the issue for ascii characters because it seems UriComponentsBuilder does not encode anything else.
I'm not sure if there is an easy solution. I have seen your PR (#3437), which would encode space and the en dash correctly. But for rewrite path it would do the regex on the encoded url which i guess no one expects when writing a rewrite rule.

@rworsnop @spencergibb
Do you have an idea how we could support both, rewrite (applying regex) on the correct path and still maintain support for non-ascii?

[jensmatw]

@rworsnop it seems that neither jdks URI nor the UriComponentsBuilder from Spring escapes unicodes outside of ascii.

jshell> new URI("http", "localhost", "/te-st test", null)
$32 ==> http://localhost/te-st%20test

UriComponentsBuilder.fromUri(request.uri()).replacePath("/te-st test").build().toUri();
http://localhost:8080/te-st%20test

Yet, the code from PR is still working for unicode characters, because it seems the http-client or something after the filter is escaping the en-dash correctly. But I found another issue: While the URI from componentsbuilder with te-st is sending te%E2%80%93st and te set is sending te%20% there is an issue if you have both! If we leave the filter with the URI te-st test it gets send to the downstream server as te%E2%80%93st%2520test, so it is double-encoded again. Something in the spring chain is "detecting" the unicode character and decides to encode everything again. Maybe it is also somewhere in the JDK.

[jensmatw]

I've found a solution, it seems .encode() from the UriComponentsBuilder does encode everything correctly. It does not double encode the space. It still encodes the unicodes (see tests).

@spencergibb tests have also been added