cw1-whitelist-ng: Contract implementation in terms of semantical structures #499

hashedone · 2021-10-19T16:53:40Z

Closes #494

hashedone · 2021-10-20T12:04:04Z

Couple comments from my side as an introduction (I feel as it is neccessary for +2k lines PR).

First of all there are some files which can be probably ignored - NOTICE, README.md, schemas are taken straight from cw1-whitelist contract. Also I strongly encourage to make review commit by commit - no much changes happened in the process, additions only (besides some little alignments), and they are just incremental upgrades.

Now about implementation. It is obvious there is a lot, even more than in default implementation. However most of this code is to be generated by the attribute macro. Therefor things which would land in the final contract, would be:

interfaces.rs - you can look at this as on an equivalent of what was before in msg.rs. Using traits in place of rigid msg structures would allow to implement multiple independent traits for same contract (as for eg. any contract reacting to cw20 message can implement its own trait + Cw20Receiver trait). Code generation takes care about generating proper messages, dispatching functions and entry points, so at the end underlying API is just exactly the same as before
contract.rs which has exactly the same purpose as before, the difference is that instead of implementing raw functions, everything is just implementing traits (or methods for migrate and instantiate, how exactly sudo and reply would be handled is an open question I am working out now - I have some ideas here). The generated dispatching function would take care to call proper functions for every messages. Note, that entry points disappeared from here - they are moved directly to lib.rs, and are generated only for non-library target (there are nevermore needed for multitest - will be explained in the moment)
state.rs which also didn't change its semantics, but instead of global statics, state is encapsulated in structure. This would work better with extending contracts, and with things with non-const constructors (as multi index maps). If it is possible to create constant constructor here, it is provided and contract is kept as constant on execution, as it was before, so no additional overhead here

Now a word about new things. There are two new files to be generated:

query.rs is an utility to perform smart query with direct interfaces, as similar to the raw interfaces as it is possible. I am considering adding similar thing for generating execution sub messages, but I am not convinced that it would simplify anything.
multitest/contract.rs is an implementation of multitest Contract directly on the contract type, so no ContractWrapper is necessary for multitest. It has additional advantage for contracts where state is not static const - the state is build once and kept (not a big deal in tests, but always nice). What is more important - the additional proxy type is generated for easier and fully typed execution on contract (with access to queries as well), so multitests would be even more nice to read. Note, that in this example I resigned from Suite approach, but it would still be nice for most bigger cases (mostly to keep test initialization compact, maybe to add wrappers to cut obvious fields, like send_funds.

In terms of tests itself - don't expect changes there. Only alignments to slight API changes (like calling functions on object instead of global functions). Whole logic is exactly the same.

ethanfrey · 2021-10-20T13:46:24Z

schemas are taken straight from cw1-whitelist contract

Time to remove all schemas!!! @maurolacy was excited by this one :)

ethanfrey · 2021-10-20T13:48:22Z

Starting review now... I'm excited 😁

ethanfrey

Nice work. It looks clean and usage looks simple (in the tests).

I added a number of comments on how to clean up a few points here and there but the general design looks good.

I am curious how you plan to auto-generate the Msgs, but looking forward to those surprises as well.

ethanfrey · 2021-10-20T13:49:39Z

contracts/cw1-whitelist-ng/src/interfaces.rs

+        admins: Vec<String>,
+    ) -> Result<Response<T>, Self::Error>;
+
+    fn admin_list(&self, deps: Deps, env: Env) -> Result<AdminListResponse, Self::Error>;


Okay, so one trait for both execution and queries.
Curious if this will need to be separated later on when we mock out / customize, but happy to start with this.

I think it would be one trait for both queries and exec, separation would be on the other level - one trait per API. Therefor this one would probably break into:

pub trait Cw1 { fn execute(...) -> ...; fn can_execute(...) -> ...; } pub Whitelist { fn freeze(...) -> ...; fn update_admins(...) -> ...; fn admin_list(...) -> ...; }

The Cw1 would be eventually extracted to packages/cw1.

That would be cool to see, even in the same package. Just to show how extension works.

pub trait Whitelist: Cw1 {} is the same pattern we would need for one contract extending another one, same as multiple with the same interface, etc. We get the issue how how to dispatch messages.

Cw1Message and Cw1Query that can be dispatched to Cw1 trait, and somehow embed that in a WhitelistMsg that can dispatch to either Whitelist or Cw1... not sure if I am clear. Anyway, doing some (simple) trait extension will open up a large design space that is worthwhile to explore

I completely don't see a reason why Whitelist would have any relation to Cw1. I can have whitelist restricting any other contract, those are two separate ideas.

Fair enough. No need to have the traits extend each other

Let's get it to work now as one trait.
Having it as two separate traits and showing how the messages work with that would be a great demo to show composition. This can be a future step, just see how you can demo that even in one contract.

I think having different traits for execution and queries would still be good, by example for auto-generation.

How do you plan to mark / detect queries from execution? Using macro attributes? Isn't there another / a better option?

contracts/cw1-whitelist-ng/src/state.rs

contracts/cw1-whitelist-ng/src/contract.rs

contracts/cw1-whitelist-ng/src/state.rs

contracts/cw1-whitelist-ng/src/contract.rs

ethanfrey · 2021-10-20T16:23:59Z

Thanks for the responses.

hashedone · 2021-10-21T14:24:16Z

Another iteration, mostly addressing @ethanfrey comments.

Most important changes:

I actually split interfaces to Cw1 one and Whitelist one. It makes more clear how to proper dispatch on different messages, and also how API of multitest proxy should exactly look like
Entry points are now included back directly into contract implementation, as methods. The reason is, that once I included dispatching between different messages, it turns out that there is bunch of logic there, and you want to ensure that it works. Without that it may be a case, that multitest tests are working, but it doesn't work with actual contract. The actual generated entry points are basically just forwards to those methods on contract.
Removed the split between bound and unbound querier. Especially after split of interfaces it seems unlikely to store those directly as it is very limiting, and building querier inplace doesn't seems to be this much an issue.
Altered contract helpers for mutlitests. Basically made it more granular, more code, but I think simple code, which should improve how it is used in multitest (it is visible in MT example).

ethanfrey

Good stuff.

The only open comments are minor details, please merge this and we can address those with a much smaller diff.

how parse errors are displayed in enrtry_execute when failing to match on all enums.
naming of how to pass contract args to the instantiator in multitest

Like I said, minor cosmetics. But this design is awesome! And I can imagine much of this disappearing into macro-land.

Last comment - to not confuse other people, maybe we move this to contracts-ng/cw1-whitelist-ng or something... so the different styles of contracts are not next to each other? Not now, but maybe in a future PR... this is mainly me trying to think how to support two versions of the same logic in different formats without confusing the consumers.

ethanfrey · 2021-10-21T14:38:45Z

contracts/cw1-whitelist-ng/src/contract.rs

+    where
+        T: DeserializeOwned,
+    {
+        let mut errs = vec![];


Interesting....

So it tries to parse into each enum and returns all the error messages appended.
I would be curious to see a test that asserted the error string when some garbage like {"foo": "bar"} was passed in. (I know, asserting error strings is bad, maybe doc test?? something to show this doesn't look too terrible).

Minor point - can be a follow up PR.

I am struggling here. Before it was done by just dumping as an error the error message (not even ParseErr was constructed). I would actually prefer to return error looking like:

{ "parse_err": { "Cw1ExecMsg": "...", "WhitelistExecMsg": "...", } }

This actually can be done by providing additional custom error type which is either DispatchErr(HashMap<String, StdError>) (this very case) or ContractErr(Self::Error) (for failing in contract after dispatching succeed. For now I think this would be the best approach, but didn't want to put it here yet as I am polishing this idea.

ethanfrey · 2021-10-21T14:40:17Z

contracts/cw1-whitelist-ng/src/contract.rs

+    {
+        let mut errs = vec![];
+
+        match from_slice::<Cw1ExecMsg<T>>(msg) {


This multiple attempts to deserialize is a nice approach. It looks a little ugly here, but I see the beauty of each message being able to implement dispatch and the contract being able to implement multiple traits.

If this is auto-generated (which it will be shortly), then I definitely see the bonus.

I don't like this approach of "trying until succeeding", at all. It's inefficient, and unclear. It's also untyped... you're introducing a lot of types / encapsulation, and that's breaking badly here.

Isn't there a way to annotate the type a msg belongs to, and / or to use (a series of) typed handlers / pre-dispatchers?

It is as efficient as it can be - and basically what serde does. It is what all parsers do basically - they check which pattern mach and choose it, and if at some point parsing stucks - it performs fallback to last place where it had another options. There is no other way as it to decide if message is Cw1ExecMsg other than try to parse it. And my contract reacts to more than one type of message - I need to try work on both of those.

Actually I encourage you to generate the serde::Deserialize for WhitelistExecMsg (as it has more than one variant). I assume you expect something like:

match variant(token) { "freeze" => deserialize_freeze(...), "update_admins" => deserialize_update_admins(...), _ => raise_error(...) }

But what you actually get is:

if is_freeze_variant(token) { return deserialize_freeze(...); } if is_update_admins(token) { return deserialize_update_admins(...); } raise_error(...)

Except it is way more obscured (and this is the reason why serde is not the fastest json parser for rust btw - it is the most versatile). Any way my point is - it is state of the art for the parsers to try paths unless you find one (you can call it dynamic programming, you can call it DFS graph search - it is whole the same) as it is the most generic way to approach this. I could actually use the additional final structure and make it untagged probably: https://serde.rs/enum-representations.html#untagged - I will explore this option, but it is exactly what you would get at the end (except hidden in serde generated code).

I completely do not understand how is it untyped? I am not using any type conversions or so, I take what I got from stream and try to parse it into every path I can handle, but every try is fully typed. i completely do not understand the "Isn't there a way to annotate the type a msg belongs to, and / or to use (a series of) typed handlers / pre-dispatchers?" - all those messages belongs to many contracts. I can have arbitrary number of contracts reacting on Cw1ExecMsg - all of them would be annotated with proper attribute, so dispatcher would be generated.

I explored it to figure out why I didn't use untagged - it is not working with serde_json_wasm. It is because it is relying on deserialize_any, which is not implemented (which is a shame, and as @ethanfrey is owning it I think we could add it as it is done in normal non-wasm implementation to be fully compliant with it). But at the end it is exaclty what happens in non-wasm implementation

I completely do not understand how is it untyped?

I mean, in the original version we get these messages already deserialized for us.

So, what you are saying is that the "trial and error" is being hidden from us during pre-processing? And that that's why I like it more? I'm not so sure, but, it may be. Perhaps implementing and using deserialize_any is the way to go here then. To try and hide this "ugliness".

Can't you instruct serde to parse to a family (or series) of types? That way, intermediate / partial results of the parsing can be re-utilized, instead of starting from scratch every time.

I grant you that I don't know how serde (or any other similar parser) works internally when deserializing, but it stretches my mind to think that it doesn't try to optimize / re-utilize intermediate / partial results somehow.

Given the existing state of serde-json-wasm, and our dependence on it (which is a much bigger question and out of scope here), there is no better way.

CosmWasm/serde-json-wasm#20 proposed #[serde(flatten)], which would produce a nice way to make embedded enums. Here was an attempt, but again, required serde-json, which doesn't compile to valid wasm (imports floats): #84

I would keep this code as is, and potentially make a macro to auto-generate this, or use #[serde(flatten)] after implementing that upstream.

I would envision something like:

pub enum ExecuteMsg { #[serde(flatten)] Cw1(Cw1ExecuteMsg), #[serde(flatten)] Whitelist(WhitelistExecuteMsg), }

or

#[flatten_nested_enums] pub enum ExecuteMsg { Cw1(Cw1ExecuteMsg), Whitelist(WhitelistExecuteMsg), }

which would only handle the case for enums, where all variants are enums, and likely be much easier to implement as a custom macro.

ethanfrey · 2021-10-21T14:41:35Z

contracts/cw1-whitelist-ng/src/contract.rs

+
+        assert_eq!(
+            res.messages,
+            msgs.into_iter().map(SubMsg::new).collect::<Vec<_>>()


This chain could make a nice helper function (in cosmwasm-std or cw-plus).
Or not... just a thought

We could. But it is very out of scope of this one ;)

ethanfrey · 2021-10-21T14:41:55Z

contracts/cw1-whitelist-ng/src/contract.rs

+    }
+
+    #[test]
+    fn execute_custom_messages_works() {


Thanks for adding this test to show T is properly supported

ethanfrey · 2021-10-21T14:42:19Z

contracts/cw1-whitelist-ng/src/interfaces.rs

+use cosmwasm_std::{CosmosMsg, Deps, DepsMut, Env, MessageInfo, Response};
+use cw1::query::CanExecuteResponse;
+
+pub trait Cw1<T> {


Great to show support for multiple interfaces.

I like this. If I'm not mistaken, this is the core of your idea. It's a great idea for re-usability and composition.

ethanfrey · 2021-10-21T14:45:28Z

contracts/cw1-whitelist-ng/src/query.rs

+use crate::msg::{AdminListResponse, Cw1QueryMsg, WhitelistQueryMsg};
+
+#[must_use]
+pub struct Cw1Querier<'a, C>


Nice implementation of Queriers. Look simpler and cleaner.

ethanfrey · 2021-10-21T14:47:26Z

contracts/cw1-whitelist-ng/src/multitest/contract.rs

+
+    pub fn execute<C>(self, msgs: Vec<CosmosMsg<C>>) -> AnyResult<AppResponse>
+    where
+        C: Clone + std::fmt::Debug + PartialEq + JsonSchema + Serialize + 'static,


This should be CustomMsg type now (maybe we need to update in cw-plus now this has landed in cosmwasm-std)

Nice use of late binding of where clause.

I missed its existence there, will change.

ethanfrey · 2021-10-21T14:48:23Z

contracts/cw1-whitelist-ng/src/multitest/contract.rs

+}
+
+impl<'a, App> WhitelistExecutor<'a, App> {
+    pub fn new(addr: Addr, app: &'a mut App, sender: Addr, send_funds: &'a [Coin]) -> Self {


This is a nice pattern. Always has the same XyzExecutor.new() args, then contract-specific function call.

You should not even call executor directly. You should have your Proxy after instantiation and you call:

contract.whitelist_exec(...).update_admins(...).unwrap();

ethanfrey · 2021-10-21T14:50:55Z

contracts/cw1-whitelist-ng/src/multitest/contract.rs

+        self
+    }
+
+    pub fn with<C>(self, admins: Vec<String>, mutable: bool) -> AnyResult<Cw1WhitelistProxy>


I don't like this name. The other optional with_{admin, label} methods are great.

This should have a more different name. Maybe do? I'm sure you can come up with something better...

let contract = Instantiator::new(....).with(...).unwrap(); looks a bit weird to me,.

I don't like do to much. I see your point, I somehow agree. Maybe call? perform? with_args? I am struggling. Maybe someone have an idea?

perform? with_args?

Both seem fine with me. Again, let's merge and discuss that in a separate thread

ethanfrey · 2021-10-21T14:52:46Z

contracts/cw1-whitelist-ng/src/multitest/contract.rs

+        self.0.clone()
+    }
+
+    pub fn instantiate<'a, App>(


Hmmm... so I would really call it.

let contract = Cw1WhitelistProxy::instantiate(...).with(...).unwrap();

That makes more sense, but still... I could imagine a better name.
Not going to block merge with it, but happy for more naming feedback before this name gets ported to tons of auto-generated code.

hashedone · 2021-10-21T15:31:05Z

@ethanfrey thanks for aproval, I am stalling a little with merging, as I would like give a chance for others to comment (in particular maybe some better idea about naming).

hashedone · 2021-10-21T16:41:38Z

Or as I am thinking about it... there are no so bad things there, so I am merging to not asking for reaproval of this monster on every rebase. Tomorrow morning I would make a PR with minor changes (in particular introduce CustomMsg), and there would be good place to think about better naming without rush.

maurolacy

Looks good! Just some comments / questions on naming and types. Will continue viewing it tomorrow.

maurolacy · 2021-10-21T21:16:36Z

contracts/cw1-whitelist-ng/src/lib.rs

+pub mod state;
+
+#[cfg(not(feature = "library"))]
+mod binary {


This name does not make sense to me. Perhaps wrapper or entry_points?

It is made as it is for "binary" release. entry_points is good to. I would actually get rid of it at all on auto-gen code, I did it for saving on cfg attributes - everything is anyway extracted to lib namespace below.

maurolacy · 2021-10-21T21:20:14Z

contracts/cw1-whitelist-ng/src/interfaces.rs

+use cosmwasm_std::{CosmosMsg, Deps, DepsMut, Env, MessageInfo, Response};
+use cw1::query::CanExecuteResponse;
+
+pub trait Cw1<T> {


I like this. If I'm not mistaken, this is the core of your idea. It's a great idea for re-usability and composition.

maurolacy · 2021-10-21T21:46:42Z

contracts/cw1-whitelist-ng/src/state.rs

+
+impl Cw1WhitelistContract<Empty> {
+    // Native form of this contract as it is to be created in entry points
+    pub const fn native() -> Self {


Is native a proper name here? I understand the idea, that without custom messages you are basically "stuck" in the native chain, but still, it's a little confusing. Perhaps stock or base are better names?

I don't know, it is just native type for this contract. I don't see how it is base as it is not extensible in any way (it is actually the less extensible form), it is only to build entry points. It exists basically only to make it easier to construct it in MT/UT and I actually like the naming here. stock may also work, but I have no recall seeing it anywhere in similar context and I like to reuse common naming so it is easier to get used to it.

maurolacy · 2021-10-21T21:57:50Z

contracts/cw1-whitelist-ng/src/interfaces.rs

+        admins: Vec<String>,
+    ) -> Result<Response<T>, Self::Error>;
+
+    fn admin_list(&self, deps: Deps, env: Env) -> Result<AdminListResponse, Self::Error>;


I think having different traits for execution and queries would still be good, by example for auto-generation.

How do you plan to mark / detect queries from execution? Using macro attributes? Isn't there another / a better option?

hashedone · 2021-10-22T08:18:16Z

@maurolacy - yes, by the attributes. I actually started some implementation some time ago (I will copy and continue work on it soon), an example Cw1 interface would look like this:

#[cw_derive::interface]
trait Cw1<#[custom] T> {
  type Error;
  
  #[msg(exec)]
  fn execute(&self, ctx: ExecCtx, msgs: Vec<CosmosMsg<T>>) -> Result<Response, Self::Error>;
  
  #[msg(query)]
  fn update_members(&self, ctx: QueryCtx, sender: String, msg: CosmosMsg<T>) -> Result<Response, Self::Error>;
}

You see the attribute #[msg] which describes details of message dispatching (message type is mandatory, but it is possible to add additional things there if there would be a need). Other things you see here is global cw_derive::interface (I picked cw_derive as a crate name, but don't stick to it yet), which is top-level macro for generating things from interface traits. The ExecCtx and QueryCtx are any types (also user defined) which implements proper From (for execute it would be From<(DepsMut, Env, MessageInfo)>, for query: From<(Deps, Env)> - it is just a way to reduce number of arguments in function and simplify generation slightly. But there is no reason not to put there the tuple itself.

Splitting interfaces between query and execution is just wrong, because of their semantics. You never want to implement just one of them - if contract implements Cw1 you may assume it implements whole interface, so it is ensured by enforcing you in compile time to implement whole trait. Especially in library development there is one very important rule - API and semantics first, then you figure it out how to achieve your nice and semantically proper API (even if it costs you a little bit more fancy backbone).

Actually if I would like to be like really fancy I could distinguish the execs from queries by their arguments (one take Deps, other one DepsMut, queries doesn't care about MessageInfo - but it would be way more difficult, and actually on my playing around I figured out it would be less flexible for end-user.

maurolacy · 2021-10-22T08:27:14Z

The ExecCtx and QueryCtx are any types (also user defined) which implements proper From (for execute it would be From<(DepsMut, Env, MessageInfo)>, for query: From<(Deps, Env)> - it is just a way to reduce number of arguments in function and simplify generation slightly. But there is no reason not to put there the tuple itself.

I for one prefer the explicit arguments (the tuple itself) , and don't like the Ctx stuff.

Splitting interfaces between query and execution is just wrong, because of their semantics. You never want to implement just one of them - if contract implements Cw1 you may assume it implements whole interface, so it is ensured by enforcing you in compile time to implement whole trait. Especially in library development there is one very important rule - API and semantics first, then you figure it out how to achieve your nice and semantically proper API (even if it costs you a little bit more fancy backbone).

Makes sense.

Actually if I would like to be like really fancy I could distinguish the execs from queries by their arguments (one take Deps, other one DepsMut, queries doesn't care about MessageInfo - but it would be way more difficult, and actually on my playing around I figured out it would be less flexible for end-user.

It will also be more brittle / obscure. I think macro attributes are good, both in terms of clarity / flexibility, and simplicity.

Besides, it's still too early for our own CosmWasm SmartContracts™️ DSL I guess. ;-)

maurolacy · 2021-10-22T08:41:37Z

OK, did a quick comparison of "classic" vs. "ng" (optimized) contracts (didn't check, but I assume functionality is currently equivalent):

$ ls -l artifacts/cw1_whitelist*.wasm
-rw-r--r-- 1 root root 218117 Oct 22 10:21 artifacts/cw1_whitelist_ng.wasm
-rw-r--r-- 1 root root 189938 Oct 22 10:21 artifacts/cw1_whitelist.wasm
$

"ng" contracts are ~15% larger than "classic" ones. Not a big deal, but let's keep an eye on this.

hashedone · 2021-10-22T08:44:32Z

@maurolacy - good point. It would be nice to check how they differ in terms of gas usage. And also it is interesting what makes them actually bigger - I actually think, that it is possible that the helpers like querier might do the difference, I would hide them behind feature flag.

maurolacy · 2021-10-22T09:15:04Z

It would be nice to check how they differ in terms of gas usage.

Created #505 for it.

hashedone marked this pull request as draft October 19, 2021 16:53

hashedone requested review from orkunkl, ethanfrey, maurolacy and ueco-jb October 20, 2021 12:04

hashedone marked this pull request as ready for review October 20, 2021 12:04

hashedone requested a review from uint October 20, 2021 12:21

hashedone changed the title ~~cw1-subkeys-ng: Basic implementation~~ cw1-whitelist-ng: Basic implementation Oct 20, 2021

hashedone force-pushed the 494-cw1-whitelist-ng branch from 6b0684f to 5aee758 Compare October 20, 2021 13:23

hashedone changed the title ~~cw1-whitelist-ng: Basic implementation~~ cw1-whitelist-ng: Contract implementation in terms of #391 Oct 20, 2021

hashedone changed the title ~~cw1-whitelist-ng: Contract implementation in terms of #391~~ cw1-whitelist-ng: Contract implementation in terms of semantical structures Oct 20, 2021

ethanfrey reviewed Oct 20, 2021

View reviewed changes

hashedone added 9 commits October 21, 2021 16:25

cw1-whitelist-ng: Basic implementation

8f1973a

cw1-whitelist-ng: Entry points

15af414

cw1-whitelist-ng: Helpers for querying contracts

8e0a067

cw1-whitelist-ng: Contract implementation for multitest

67c9f3f

cw1-whitelist-ng: Proxy for multitests

934a61f

cw1-whitelist-ng: Multitests ported

3707fc8

cw1-whitelist-ng: UT ported

d971dfd

cw1-whitelist-ng: Schema generation

94afab1

cw1-subkeys-ng: Improved some splitting

c0f8aea

hashedone force-pushed the 494-cw1-whitelist-ng branch from b7ca929 to c0f8aea Compare October 21, 2021 14:25

cw1-subkeys-ng: Building contract as native in entry points

1292e02

ethanfrey approved these changes Oct 21, 2021

View reviewed changes

hashedone merged commit 2851fcc into main Oct 21, 2021

hashedone deleted the 494-cw1-whitelist-ng branch October 21, 2021 16:41

maurolacy reviewed Oct 21, 2021

View reviewed changes

cw1-whitelist-ng: Contract implementation in terms of semantical structures #499

cw1-whitelist-ng: Contract implementation in terms of semantical structures #499

Conversation

hashedone commented Oct 19, 2021

hashedone commented Oct 20, 2021 • edited Loading

ethanfrey commented Oct 20, 2021

ethanfrey commented Oct 20, 2021

ethanfrey left a 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

maurolacy Oct 21, 2021 • edited Loading

Choose a reason for hiding this comment

ethanfrey commented Oct 20, 2021

hashedone commented Oct 21, 2021

ethanfrey left a 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

maurolacy Oct 22, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

maurolacy Oct 22, 2021 • edited Loading

Choose a reason for hiding this comment

maurolacy Oct 22, 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

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

Choose a reason for hiding this comment

hashedone commented Oct 21, 2021

hashedone commented Oct 21, 2021 • edited Loading

maurolacy left a comment

Choose a reason for hiding this comment

maurolacy Oct 21, 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

maurolacy Oct 21, 2021 • edited Loading

Choose a reason for hiding this comment

hashedone commented Oct 22, 2021 • edited Loading

maurolacy commented Oct 22, 2021 • edited Loading

maurolacy commented Oct 22, 2021

hashedone commented Oct 22, 2021 • edited Loading

maurolacy commented Oct 22, 2021 • edited Loading

hashedone commented Oct 20, 2021 •

edited

Loading

maurolacy Oct 21, 2021 •

edited

Loading

maurolacy Oct 22, 2021 •

edited

Loading

maurolacy Oct 22, 2021 •

edited

Loading

maurolacy Oct 22, 2021 •

edited

Loading

hashedone commented Oct 21, 2021 •

edited

Loading

maurolacy Oct 21, 2021 •

edited

Loading

maurolacy Oct 21, 2021 •

edited

Loading

hashedone commented Oct 22, 2021 •

edited

Loading

maurolacy commented Oct 22, 2021 •

edited

Loading

hashedone commented Oct 22, 2021 •

edited

Loading

maurolacy commented Oct 22, 2021 •

edited

Loading