PathParam URL encoded in quarkus-resteasy-reactive since quarkus 3.2.x

[rubensa]

Describe the bug

I have a Resource like:

@Path("/base/{name}/objects/{key:.*}/func")
public class BaseResource {

  @GET
  @Produces({ "application/json" })
  public Response runFunc(@PathParam("name") String name,@PathParam("key") String key) {
  }
}

When using quarkus-resteasy-reactive if the Resource is called with a URL like:

/base/name1/objects/path/to/my/object/func

the content of the key PathParam is:

path%2Fto%2Fmy%2Fobject

instead of:

path/to/my/object

This is happening in quarkus 3.2.5 and 3.3.2 but is working in quarkus 2.16.10.Final, 3.0.4 and 3.1.3.Final.

On quarkus 3.2.5 if I use quarkus-resteasy (no reactive) it is also working as expected (so looks like the way to parse the PathParams is different).

NOTE: I also tried adding @Encoded annotation but the result is the same.
NOTE: This bug might be related but I can't find the match among RESTeasy code and quarkus-resteasy-reactive code.

Expected behavior

To not encode the PathParam.

Actual behavior

The PathParam was encoded.

How to Reproduce?

Look at the sample code above.

Output of uname -a or ver

Linux nx-links.local.computer 6.2.0-32-generic #32~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Fri Aug 18 10:40:13 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

Output of java -version

openjdk version "17.0.8" 2023-07-18 OpenJDK Runtime Environment Temurin-17.0.8+7 (build 17.0.8+7) OpenJDK 64-Bit Server VM Temurin-17.0.8+7 (build 17.0.8+7, mixed mode, sharing)

GraalVM version (if different from Java)

No response

Quarkus version or git rev

3.2.5 and 3.3.2

Build tool (ie. output of mvnw --version or gradlew --version)

Apache Maven 3.9.4 (dfbb324ad4a7c8fb0bf182e6d91b0ae20e3d2dd9)

Additional information

No response

[quarkus-bot]

/cc @FroMage (resteasy-reactive), @geoand (resteasy-reactive), @stuartwdouglas (resteasy-reactive)

[rubensa]

#30577 might be related to this.

[rubensa]

#24974 might also be related to this.

[geoand]

@rubensa can you share a sample application with a corresponding test that behaves as you describe?

That would be super useful in pinpointing the exact version in which the change started to manifest.

[rubensa]

Here I prepared a sample project.

You can use it via VSCode with the Remote Development extension pack and Docker, via GitHub Codespaces or locally if you have GraalVM installed (in the provided Docker container for VSCode and Codespaces, Mandrel 23.0 for Java 17 is used).

As you can see in the README:

The code is configured to use Quarkus 3.4.1 version with quarkus-resteasy-reactive.

I you run the provided test it (running ./mvnw test) it will fail.

You can check that it works if you replace quarkus.platform.version from 3.4.1 to 3.1.3.Final in pom.xml.

It will also work if you keep quarkus.platform.version as 3.4.1 but use the non-reactive quarkus-resteasy by replacing quarkus-resteasy-reactive with quarkus-resteasy and quarkus-resteasy-reactive-jackson with quarkus-resteasy-jackson in pom.xml.

Setting up the sample project I just realized that it does not only "URL encodes" the PathParam but the match for the key in the @Path is not correctly resolved.

In the sample above, If you request /base/quarkus.test.sample/objects/data/test.txt/func, name is correctly resolved to quarkus.test.sample but keyis incorrectly resolved to base%2Ftest.txt/func/base/quarkus.test.sample/objects/data%2Ftest.txt (where it should have been resolved to data/test.txt or base%2Ftest.txt if you assume the encoding problem).

Just let me know if you need anything else.

[geoand]

Thanks a lot.

I'll be mostly off next week, so I'll likely check this the week after.

[rubensa]

Hi @geoand

Did you have some time to check this? 🤔

[geoand]

Hi,

Unfortunately something that requires my ful and individed attention for the few weeks (at least) has come up, so other things have been pushed down.

[rubensa]

Thanks for the info.

Hopes you have some time to check in the near future.

[rubensa]

Hi @geoand

Any news on this? For now I'm using quarkus-resteasy but I'd like to go back to use quarkus-resteasy-reactive.

[geoand]

I have been doing some high priority stuff lately and have not had time to look into his yet.
Hopefully soon.

[rmartinc]

We are affected by this one also in keycloak (issue keycloak/keycloak#25111). I think the reason is this PR: #34809. In case the regex is custom (like .* in this case), the path param is not decoded and returned with the percent encoded characters. One workaround is using PathSegment which is re-encoded again.

[geoand]

I have not yet looked at it unfortunately.

@rmartinc if you have a general fix in mind, by all means open a PR :)

[rmartinc]

@rmartinc if you have a general fix in mind, by all means open a PR :)

@geoand I just sent PR #37513. The idea is saving the path params encoded instead of decoded. That way the initial issue #34586 can be fixed doing the decoding after the split.

[geoand]

Thanks!