WIP: Test gzip support in RESTEasy Reactive #16924

saumya1singh · 2021-04-30T10:19:34Z

netodevel

@SaumyaSingh1,
I suggested some small changes.

netodevel · 2021-04-30T19:11:50Z

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

+                    .addClasses(BeanRegisteringRouteUsingObserves.class));
+
+    @Test
+    public void test() throws Exception {


method nomenclature is too generic.
If you can make it clear what is being tested

netodevel · 2021-04-30T19:13:08Z

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

+public class GZipTest {
+
+    private static final String APP_PROPS = "" +
+            "quarkus.http.enable-compression=true\n";


what do you think of a test with the other way?

quarkus.http.enable-compression=false

If it is disabled, should we really perform any sort of test? I don't think that's needed when is not even enabled :)

but how to ensure that the property is working correctly?
and how do you make sure that no future code from other developers doesn't affect this and go undetected because not have any tests?

@FroMage WDYT

Well, we should test the default behaviour, that's true, but because we know what the default is ATM, we want to test the default behaviour without any configuration, to make sure it won't be compressed by default.

FroMage · 2021-05-04T09:25:34Z

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

+
+        public void register(@Observes Router router) {
+
+            router.route("/compression").handler(rc -> {


This is testing reactive routes, but we want to be testing RESTEasy Reactive :)

From available articles (https://quarkus.io/guides/reactive-routes) I got the reactive route part, I tried searching for resources around testing RR but didn't get. Maybe we are supposed to do something like ?

quarkus/extensions/resteasy-reactive/quarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/server/test/resource/basic/SubResourceRequestFilterTest.java

Line 61 in 85d574e

@ApplicationScoped

oh wait, perhaps I should be testing this against some RR endpoint

Yes, indeed.

FroMage · 2021-05-04T09:26:30Z

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

+                .header("content-length", Matchers.not(Matchers.equalTo(Integer.toString(longString.length()))))
+                .body(Matchers.equalTo(longString));
+
+        RestAssured.given().get("/nocompression").then().statusCode(200)


For good measure, we should probably also test that we can send compressed data to the server.

So here we want to test if the data we are sending is compressed or not ( i.e Inputstream is gzipped or zipped ). Right?

Or do we want to compress the files first before sending them? (but there will be no purpose of enabling gzip then, so not this one)

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

…ent/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java Co-authored-by: Stéphane Épardaud <[email protected]>

Registering endpoints

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

saumya1singh · 2021-05-10T07:09:39Z

@FroMage
#16425 (comment)

In order to automate the process of "stoping the compression" we want to create an annotation.

Situations/conditions when we want to stop/disable compression -

If the data/files are already compressed (can be checked by magic number concept, .zip or .gz begins with specific magic num. OR by extension type?)
If there are images (.png or .gif ) then we want to stop compression

FroMage · 2021-05-07T13:10:26Z

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

+                .header("content-encoding", "identity")
+                .header("content-length", Matchers.equalTo(Integer.toString(longString.length())))
+                .body(Matchers.equalTo(longString));
+    }


So does this test pass? I don't see anything in the endpoints that would justify a different behaviour.
Could you test with returning a binary reponse, such as an image? Just to see what vert.x does in this case, if it stll tries to compress it.

It is failing with this info.
For endpoint /test/nocompression

I created an image and returned to the endpoint. This time also, content-encoding is gzip.

So in the beginning I am initializing APP_PROPS as true

private static final String APP_PROPS = "" + "quarkus.http.enable-compression=true\n";

perhaps that's why content encoding is always gzip even for Binary Data i.e Images

That means vert.x is trying to compress the binary data as well, no?

When I set quarkus.http.enable-compression to false , the content-length=24000 bytes which is far more than 130 bytes (for text) or 165 bytes(for image)

private static final String APP_PROPS = "" + "quarkus.http.enable-compression=false\n";

So perhaps the conclusion is - It is trying to compress images as well.

Am I right @FroMage?

Yes, probably, could you add that test to your PR?

Yeah sure, I missed this comment.

FroMage · 2021-05-10T09:40:16Z

Situations/conditions when we want to stop/disable compression -
If the data/files are already compressed (can be checked by magic number concept, .zip or .gz begins with specific magic num. OR by extension type?)
If there are images (.png or .gif ) then we want to stop compression

Well, first I'd like to see you test what the current behaviour is: will binary types get compressed or not?

saumya1singh · 2021-05-10T12:22:28Z

Well, first I'd like to see you test what the current behaviour is: will binary types get compressed or not?

Seems yes.

For images :

When quarkus.http.enable-compression is false ->

When quarkus.http.enable-compression is true ->

FroMage · 2021-05-10T13:34:16Z

Since we can't enable compression on a per-endpoint basis, I guess we really need to add a @DisableCompression annotation, and make RESTEasy Reactive automatically add the header as an extra handler.

saumya1singh · 2021-05-10T14:04:21Z

Interesting, so now I am supposed to create an annotation @DisableCompression.
This annotation should automatically add the header Transfer-Encoding: identity to stop compression.

And we will add this annotation at the endpoint test/nocompression.

FroMage · 2021-05-10T14:21:15Z

Yes.

saumya1singh · 2021-05-11T02:58:30Z

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

+
+        RestAssured.given().get("/test/nocompression").then().statusCode(200)
+                .header("content-encoding", "identity")
+                .header("content-length", Matchers.equalTo((long) (createImage().getData().getDataBuffer().getSize() * 4L)))


(long) (createImage().getData().getDataBuffer().getSize() * 4L)

Finding the content-length for BufferedImage is something I am not much sure about.
I took reference from here -
https://stackoverflow.com/questions/632229/how-to-calculate-java-bufferedimage-filesize?noredirect=1&lq=1

The size of this png image(original) is 2.91kb OR 2910 bytes

Using the above technique, the size is 360000 bytes.

On calculating content-length using ByteArrayOutputStream I am getting 2914 bytes. ✔️
So perhaps this is the correct techniques to get the byte size of BufferedImage :

BufferedImage image = createImage(); ByteArrayOutputStream outputStream = new ByteArrayOutputStream(); ImageIO.write(image, "png", outputStream); outputStream.close(); long contentLength = outputStream.size(); System.out.println(" content-length using ByteArrayOutputStream => " + contentLength );

Actually you can just use a static test image as a resource :)

Humm I thought it's better to generate an image only when this test is run.

You are right, it will make code simpler. Instead of creating an image through code it is better to use a static image 💯

Done.

Original size - 2914 bytes
Compressed size - 2741 bytes

saumya1singh · 2021-05-11T03:00:00Z

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java

+        ImageIO.write(bufferedImage, "png", file);
+        return bufferedImage;
+    }
+}


The image I am creating in createImage() looks like this -

Perhaps "saving the image" is optional.

saumya1singh · 2021-05-11T15:07:24Z

I have created the annotations, but I am yet to understand and implement the logic that how to make my custom annotation set the header Transfer-Encoding to identity.

Am I going in the right direction? @FroMage

FroMage · 2021-05-11T16:10:17Z

I think your annotation should live next to ServerExceptionMapper, and then you have to make EndpointIndexer.createResourceMethod understand that annotation, and pass the info that compression is disabled in ResourceMethod which is used by RuntimeResourceDeployment.buildResourceMethod in order to create the set of ServerRestHandler that are to be called for the endpoint.

So if the endpoint has compression disabled, we need a new handler that adds the required header after the endpoint is invoked, probably before this line:

        addHandlers(handlers, method, info, HandlerChainCustomizer.Phase.AFTER_METHOD_INVOKE);

BTW, I'll be on PTO the next three days so take your time, or ask others for help on Zulip if you need to, but it's possible they'll also be on PTO.

saumya1singh · 2021-05-12T03:24:19Z

Sure, it's public holiday here as well (tomorrow ig).

I will use your explanation ^ and try to understand, implement it.

saumya1singh · 2021-05-17T07:38:52Z

and then you have to make EndpointIndexer.createResourceMethod understand that annotation, and pass the info that compression is disabled in ResourceMethod

In order to do this ^ I tried thinking of these 2 ways :

1. Should I directly introduce the new annotation inside createResourceMethod() in EndPointIndexer.java

quarkus/independent-projects/resteasy-reactive/common/processor/src/main/java/org/jboss/resteasy/reactive/common/processor/EndpointIndexer.java

Line 377 in c88b990

    
           private ResourceMethod createResourceMethod(ClassInfo currentClassInfo, ClassInfo actualEndpointInfo,

and pass the info that compression is disabled somewhere in a new ResourceMethod

quarkus/independent-projects/resteasy-reactive/common/processor/src/main/java/org/jboss/resteasy/reactive/common/processor/EndpointIndexer.java

Line 500 in c88b990

    
           ResourceMethod method = createResourceMethod(currentMethodInfo, actualEndpointInfo, methodContext)

OR
2. In QuarkusServerEndpointIndexer

quarkus/extensions/resteasy-reactive/quarkus-resteasy-reactive/deployment/src/main/java/io/quarkus/resteasy/reactive/server/deployment/QuarkusServerEndpointIndexer.java

Line 50 in 2cbd177

public class QuarkusServerEndpointIndexer

override createResourceMethod() and set the properties in a way to disable compression.

saumya1singh · 2021-05-17T07:49:04Z

While working on the previous filter and other tasks I was easily able to debug my code written in DevConsoleProcessor or other files by hitting the DEv-UI endpoints such as http://localhost:8080/q/dev/io.quarkus.quarkus-resteasy-reactive/endpoints.

How should I go about it here e.g to check breakpoints in EndpointIndexer
https://github.com/quarkusio/quarkus/blob/c88b990f3163785e0eaaa59acd98db4c8657793c/independent-projects/resteasy-reactive/common/processor/src/main/java/org/jboss/resteasy/reactive/common/processor/EndpointIndexer.java

FroMage · 2021-05-18T15:21:41Z

I would say option 1, because this is not specific to Quarkus.

FroMage · 2021-05-18T15:23:43Z

Breaking on the indexer requires you to run the build steps, but you can do that by adding a QuarkusUnitTest in quarkus-resteasy-reactive-deployment like the others there, and running that test will invoke the build steps and therefore your breakpoint.

gastaldi · 2022-02-23T20:38:18Z

Hi @saumya1singh, any chance you'll resume this soon?

saumya1singh · 2022-02-24T04:20:21Z

hey @gastaldi, after 2.2.5 I am focusing on the Elektra release (plus have to provide training sessions to new interns in prod) , so can't give much time on this issue :/

gastaldi · 2022-02-24T05:27:16Z

@saumya1singh no worries, just curious 😉

mkouba · 2022-03-15T15:29:26Z

For the record - it seems that vertx will try to compress the static resources as well, see the StaticHandler.skipCompressionForMediaTypes() and StaticHandler.skipCompressionForSuffixes() methods. And I think that we should make this configurable too.

mkouba · 2022-03-15T16:22:04Z

Hm, to be honest I don't like the idea of opt-out via @DisableCompression (and config properties for static resources). I do understand that we're limited by how vertx-web implements compression but still.. it's not very flexible.

In theory, we could introduce two annotations (@Compressed and @Uncompressed) and a property to specify the default behavior for endpoints if quarkus.http.enable-compression=true, e.g. something like quarkus.resteasy-reactive.endpoints-compressed (false by default?). Furthermore, we could introduce a simple @ServerResponseFilter that would add the Transfer-Encoding: identity header if the resource method is annotated with @Uncompressed OR quarkus.resteasy-reactive.endpoints-compressed=false and the method is not annotated with @Compressed. @saumya1singh @FroMage WDYT?

gsmet · 2022-03-21T15:23:16Z

@geoand @cescoffier could we try to come up with a plan so that @mkouba can put it into motion.

As for the requirements, I think:

we should be able to compress endpoints without compressing static resources (and reciprocally)
we should be able to enable/disable compression (with a global flag and then opt-in/opt-out)
we will need to make sure gzip works for Reactive REST Client - I'm not sure if we handle GZip there at the moment

I don't think we need fancy algorithm to start with. Let's keep it simple and we will see later if we can expand.

geoand · 2022-03-21T15:27:41Z

I'll have a look (although I know nothing about how Vert.x supports gzip), but honestly I don't conside this terribly important fot 2.8.

cescoffier · 2022-03-21T16:36:12Z

Sorry, I don't have much availability this week.

First, the static handler can skip compression for a set of media types or extensions. I think we should rely on this.

Basically, I would recommend enabling compression by default and letting decide compression vs. raw using the annotation proposed by @mkouba. I know it's a big change in terms of behavior, but I think it's reasonable, and the user can go back to the previous behavior by disabling the compression globally.

So, in other words, I agree with @mkouba's plan. We would need to document how to skip compression for static files based on the media types (I'm not sure we expose these settings, so we should).

Note that compression is handled at the netty level. Basically, we add the netty handler compressing the response. That's why it's global.

gsmet · 2022-03-21T16:44:23Z

honestly I don't consider this terribly important fot 2.8

I tend to disagree with that. We had numerous reports/questions about usage of the GZip option and it's especially important if you're paying your bandwidth bills in a cloud environment.

cescoffier · 2022-03-21T17:08:10Z

It's a question of trade off as you use more cpu and so cannot handle the same level of concurrency (which means you may need more instances).

I also wanted to add the vertx 4.3 is going to extend how is handled compression. That may give us more flexibility, however that's for June (and I do not believe we can wait)

mkouba · 2022-03-22T08:03:06Z

Ok, to sum it up:

we should change the default so that quarkus.http.enable-compression=true
introduce @Compressed and @Uncompressed
introduce a config property to skip compression for static resources; and maybe we should skip ALL by default (the only problem is that StaticHandler.skipCompressionForMediaTypes() and StaticHandler.skipCompressionForSuffixes() methods accept an explicit set and don't support wildcards or anything like that)
(?) introduce a config property to enable/disable compression for a specific technology, e.g. resteasy-reactive resource methods via quarkus.resteasy-reactive.compressed-endpoints, false by default

This would allow a user to:

by default: just add @Compressed to a specific resource method and other resource methods would remain uncompressed
enable compression for all resource methods via quarkus.resteasy-reactive.compressed-endpoints=true and then opt-out via @Uncompressed where needed
enable compression for some static resources
disable the compression altogether with quarkus.http.enable-compression=false

geoand · 2022-03-22T08:09:46Z

I tend to disagree with that. We had numerous reports/questions about usage of the GZip option and it's especially important if you're paying your bandwidth bills in a cloud environment.

Not that I have seen

geoand · 2022-03-22T08:10:42Z

This would allow a user to:

by default: just add @Compressed to a specific resource method and other resource methods would remain uncompressed

enable compression for all resource methods via quarkus.resteasy-reactive.compressed-endpoints=true and then opt-out via @Uncompressed where needed

enable compression for some static resources

disable the compression altogether with quarkus.http.enable-compression=false

Seems reasonable in theory, but I don't really understand how compression can be used on per resource basis.

mkouba · 2022-03-22T09:07:59Z

Seems reasonable in theory, but I don't really understand how compression can be used on per resource basis.

Well, I can imagine a resource with methods that return large JSON payloads and also with a method foo() that returns a zip file. In this case, you'd want to quarkus.resteasy-reactive.compressed-endpoints=true and foo() annotated with @Uncompressed?

geoand · 2022-03-22T09:11:28Z

My question was mainly from an implementation point of view. So basically what you are proposing is that we have two different ways of compressing output, one general one based on Vert.x and one custom one?

mkouba · 2022-03-22T09:25:46Z

My question was mainly from an implementation point of view. So basically what you are proposing is that we have two different ways of compressing output, one general one based on Vert.x and one custom one?

Yes, but maybe I'm just overthinking on this issue ;-).

The thing is that if we merge the quarkus.http.enable-compression and quarkus.resteasy-reactive.compressed-endpoints; and interpret quarkus.http.enable-compression=true as "compress all endpoints in all technologies" (resteasy-reactive, reactive-routes) then you won't be able to disable compression for endpoints by default an opt-in via @Compressed (i.e. there's no need for this annotation). Which might be acceptable. I'm not sure.

FroMage · 2022-03-22T10:14:38Z

It could be useful to look at how web servers do it: https://httpd.apache.org/docs/2.4/mod/mod_deflate.html and https://docs.nginx.com/nginx/admin-guide/web-server/compression/

Apparently, Apache doesn't enable anything by default, but gives two documented options:

Compress html, text, js, css
Compress everything but images

Nginx by default will compress only html, but documents the option to compress also xml

From that, I gather that we could go the same way and provide an option for compression by mime types:

quarkus.http.enable-compression=text/html,text/plain,text/xml,text/css,text/javascript,application/javascript

I guess that would accept MIME wildcards too. This applies to static resources as well as endpoints. This seems to be what makes the most sense and is the most flexible, no?
Or, we make it two options:

quarkus.http.enable-compression=true
# this is the default value, when compression is enabled
#quarkus.http.enable-compression.for=text/html,text/plain,text/xml,text/css,text/javascript,application/javascript

Now, sure, we can also support an annotation to make it per endpoint, but IMO that's a "nice to have" and not as important as this sort of config.

geoand · 2022-03-22T10:59:47Z

My question was mainly from an implementation point of view. So basically what you are proposing is that we have two different ways of compressing output, one general one based on Vert.x and one custom one?

Yes, but maybe I'm just overthinking on this issue ;-).

The thing is that if we merge the quarkus.http.enable-compression and quarkus.resteasy-reactive.compressed-endpoints; and interpret quarkus.http.enable-compression=true as "compress all endpoints in all technologies" (resteasy-reactive, reactive-routes) then you won't be able to disable compression for endpoints by default an opt-in via @Compressed (i.e. there's no need for this annotation). Which might be acceptable. I'm not sure.

Yeah, this is exactly why I think we should not rush into this... It seems to me like we need to put some more thought into it.

mkouba · 2022-03-22T11:14:58Z

From that, I gather that we could go the same way and provide an option for compression by mime types:

Well, I like the idea but the problem is that StaticHandler currently does not support the opt-in approach; i.e. you can only say "enable compression and disable for a set of media types".

I guess that would accept MIME wildcards too.

I don't think the current StaticHandlerImpl supports this; it seems to use Object.equals().

@FroMage @geoand

mkouba · 2022-03-24T15:06:08Z

From that, I gather that we could go the same way and provide an option for compression by mime types:

Well, I like the idea but the problem is that StaticHandler currently does not support the opt-in approach; i.e. you can only say "enable compression and disable for a set of media types".

@cescoffier had an idea that we could add Content-Encoding: identity to every response automatically (if quarkus.http.enable-compression=true) and let the next route in the chain decide whether to remove this header (and enable compression). I've experimented with this approach and it seems to work. I will add a link to my WIP branch once the GitHub UI is in sync with my quarkus repository...

mkouba · 2022-03-24T16:05:59Z

FYI https://github.com/mkouba/quarkus/tree/issue-16425

geoand · 2022-10-18T06:16:00Z

Closing as this has been addressed in #24558

quarkus-bot bot added the area/rest label Apr 30, 2021

netodevel suggested changes Apr 30, 2021

View reviewed changes

FroMage requested changes May 4, 2021

View reviewed changes

saumya1singh and others added 3 commits May 6, 2021 12:11

gzip support in RESTEasy Reactive

bdd5a5f

Update extensions/resteasy-reactive/quarkus-resteasy-reactive/deploym…

09c5fa7

…ent/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java Co-authored-by: Stéphane Épardaud <[email protected]>

Registering endpoints

f69d0b6

Registering endpoints

saumya1singh force-pushed the issue-16425 branch from 434587d to f69d0b6 Compare May 6, 2021 06:45

saumya1singh commented May 6, 2021

View reviewed changes

...uarkus-resteasy-reactive/deployment/src/test/java/io/quarkus/resteasy/reactive/GZipTest.java Outdated Show resolved Hide resolved

Registering Endpoints

3f8f018

saumya1singh requested a review from FroMage May 7, 2021 02:26

FroMage reviewed May 10, 2021

View reviewed changes

GZip:Adding Binary Data at Endpoint

aa7bd48

saumya1singh commented May 11, 2021

View reviewed changes

saumya1singh added 3 commits May 11, 2021 13:10

Gzip:content-length for Image

e45b503

gZip: created annotation DisableCompression

2eb085a

gZip: created annotation DisableCompression

4438534

gzip : Image fix

045548d

saumya1singh changed the title ~~WIP: gzip support in RESTEasy Reactive~~ WIP: Test gzip support in RESTEasy Reactive May 13, 2021

geoand closed this Oct 18, 2022

quarkus-bot bot added the triage/invalid This doesn't seem right label Oct 18, 2022


		public void register(@Observes Router router) {

		router.route("/compression").handler(rc -> {

WIP: Test gzip support in RESTEasy Reactive #16924

WIP: Test gzip support in RESTEasy Reactive #16924

Conversation

saumya1singh commented Apr 30, 2021

netodevel left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

netodevel Apr 30, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

netodevel May 1, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

saumya1singh May 10, 2021 • edited Loading

Choose a reason for hiding this comment

saumya1singh commented May 10, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

saumya1singh May 10, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

saumya1singh May 10, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

FroMage commented May 10, 2021

saumya1singh commented May 10, 2021

FroMage commented May 10, 2021

saumya1singh commented May 10, 2021

FroMage commented May 10, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

saumya1singh commented May 11, 2021 • edited Loading

FroMage commented May 11, 2021

saumya1singh commented May 12, 2021

saumya1singh commented May 17, 2021

saumya1singh commented May 17, 2021

FroMage commented May 18, 2021

FroMage commented May 18, 2021

gastaldi commented Feb 23, 2022

saumya1singh commented Feb 24, 2022

gastaldi commented Feb 24, 2022

mkouba commented Mar 15, 2022

mkouba commented Mar 15, 2022

gsmet commented Mar 21, 2022

geoand commented Mar 21, 2022 • edited Loading

cescoffier commented Mar 21, 2022

gsmet commented Mar 21, 2022

cescoffier commented Mar 21, 2022

mkouba commented Mar 22, 2022 • edited by cescoffier Loading

geoand commented Mar 22, 2022

geoand commented Mar 22, 2022 • edited Loading

mkouba commented Mar 22, 2022

geoand commented Mar 22, 2022

mkouba commented Mar 22, 2022

FroMage commented Mar 22, 2022

geoand commented Mar 22, 2022

mkouba commented Mar 22, 2022 • edited Loading

mkouba commented Mar 24, 2022

mkouba commented Mar 24, 2022

geoand commented Oct 18, 2022

netodevel Apr 30, 2021 •

edited

Loading

netodevel May 1, 2021 •

edited

Loading

saumya1singh May 10, 2021 •

edited

Loading

saumya1singh commented May 10, 2021 •

edited

Loading

saumya1singh May 10, 2021 •

edited

Loading

saumya1singh May 10, 2021 •

edited

Loading

saumya1singh commented May 11, 2021 •

edited

Loading

geoand commented Mar 21, 2022 •

edited

Loading

mkouba commented Mar 22, 2022 •

edited by cescoffier

Loading

geoand commented Mar 22, 2022 •

edited

Loading

mkouba commented Mar 22, 2022 •

edited

Loading