Define and implement block rewards #1768

ferencdg · 2021-03-11T05:16:44Z

fixes: #246

ferencdg · 2021-03-11T05:18:33Z

source/agora/test/BlockReward.d

@@ -0,0 +1,87 @@
+/*******************************************************************************


Will add 2 more tests: missing signatures and missing validators

ferencdg · 2021-03-11T05:20:11Z

source/agora/consensus/validation/Block.d

@@ -148,13 +148,6 @@ public string isInvalidReason (in Block block, Engine engine, Height prev_height
    if (only_coinbase)
        return "Block: Must contain other transactions than Coinbase";

-    auto expected_cb_txs = getCoinbaseTX(block.txs,


point 5
#246 (comment)

ferencdg · 2021-03-11T05:20:37Z

source/agora/test/Fee.d

@@ -214,10 +214,6 @@ unittest
    {
        createAndExpectNewBlock(Height(block_idx));
    }
-


point 5
#246 (comment)

Does this mean a node can be fed blocks with wrong CB txs? Do we just trust the peer when we are catching up?

Does this mean a node can be fed blocks with wrong CB txs? Do we just trust the peer when we are catching up?

We are still checking if at least 50% of the validators signed the block, so not just trusting only 1 peer.

------ background

However, Mathias described an attack before

in 2020, lets say we have 20 validators, 2 are malicious, 8 are lazy(they don't sign blocks), and 10 are well-behaving. Those 10 well-behaving is able to create a block in 2020.

in 2022, 8 validators(from 2020) sell their private keys(6 previously lazy ones and 2 previously well-behaving one are the sellers).

now those 2 malicious validators from 2020 along with the newly bought private keys can dupe a 'catching up' a validator.

Our block reward(based on validators signing blocks) will try to minimize the number of lazy validators, so let's in 2020 we didn't have 8 lazy validators, just 2, so the block was signed by 6+10=16 validators... With block reward in mind the person in 2022 have to buy more private keys, because we will also change our catchup logic at some point to the following one:

each validator needs to accept the block with the MOST signatures at a specific height

----- end of background

so if we implement all above, then a catching up validator can only be duped, if someone can create a block with let's say 60%-70% of all signatures

ferencdg · 2021-03-11T05:28:40Z

source/agora/node/Ledger.d


        Returns:
-            List of expected Coinbase TXs
+            One or zero Coinbase TX


explanation
0cd1c01#diff-d25eeef12241e794638a748cea9cc8e7b124f1d61ccdaa962f7fb64ed3c66cb6R725

actually, even before my change we had numerous asserts that this function only returns 1 or less elements in the array

codecov · 2021-03-11T05:54:30Z

Codecov Report

Merging #1768 (da52927) into v0.x.x (6f78c68) will increase coverage by 0.67%.
The diff coverage is 96.48%.

❗ Current head da52927 differs from pull request most recent head 3055b73. Consider uploading reports for the commit 3055b73 to get more accurate results

@@            Coverage Diff             @@
##           v0.x.x    #1768      +/-   ##
==========================================
+ Coverage   88.05%   88.72%   +0.67%     
==========================================
  Files         144      147       +3     
  Lines       13869    13825      -44     
==========================================
+ Hits        12212    12266      +54     
+ Misses       1657     1559      -98

Flag	Coverage Δ
integration	`42.36% <ø> (+4.47%)`	⬆️
unittests	`88.50% <96.48%> (+0.75%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
source/agora/common/Set.d	`90.90% <ø> (-0.40%)`	⬇️
source/agora/consensus/validation/Block.d	`95.75% <ø> (-0.12%)`	⬇️
source/agora/test/Base.d	`76.66% <ø> (+2.81%)`	⬆️
source/agora/test/Fee.d	`95.65% <ø> (-0.10%)`	⬇️
source/agora/test/MissingPreImageDetection.d	`93.33% <ø> (+0.31%)`	⬆️
source/agora/test/ValidatorRecurringEnrollment.d	`92.30% <ø> (-2.14%)`	⬇️
source/agora/utils/Log.d	`63.46% <50.00%> (-8.10%)`	⬇️
source/agora/consensus/EnrollmentManager.d	`95.06% <75.00%> (-0.78%)`	⬇️
source/agora/common/Amount.d	`97.46% <80.00%> (-0.61%)`	⬇️
source/agora/node/Ledger.d	`94.13% <91.66%> (+<0.01%)`	⬆️
... and 102 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6f78c68...3055b73. Read the comment docs.

ferencdg · 2021-03-15T01:37:45Z

ready for review - squashed some commits and cleaned up the code

AndrejMitrovic

Just a few comments.

I'm not sure whether using floating point is a good idea, there were reservations against using it in the past.

AndrejMitrovic · 2021-03-15T06:21:31Z

source/agora/common/Amount.d

@@ -86,6 +86,13 @@ public struct Amount
        return ret;
    }

+    /// Non-throwing version of toString
+    public string toStringNT () const @safe nothrow


Why is this needed? Commit doesn't explain it.

Why is this needed? Commit doesn't explain it.

I had some 'nothrow' functions from which I had to print out the 'Amount', I will modify the comment to make it more obvious that this 'non-throwing' version should mainly be used for debugging.

It's not used anywhere and stands out, I'd remove it.
If you need to debug you can use the scope (failure) assert(0); hack.

AndrejMitrovic · 2021-03-15T06:24:32Z

source/agora/consensus/BlockRewardCalc.d

+       block signatures which is the basis of the block reward.
+    3. Each validator will be rewarded based on how many blocks they signed
+       and the reward will NOT be propotional to how much coin they staked.
+    4  The block reward is divided between the validators and the Foundation.


s/Foundation/Commons Budget?

AndrejMitrovic · 2021-03-15T06:25:31Z

source/agora/consensus/BlockRewardCalc.d

+
+mixin AddLogger!();
+
+public class BlockRewardCalc


Needs /// or ddoc is not generated.

AndrejMitrovic · 2021-03-15T06:27:44Z

source/agora/consensus/BlockRewardCalc.d

+    public Amount[Point] getEachValidatorReward (in Height height, ushort[Point] validator_stakes,
+        out Amount remainder, double comply_perc, uint indent = 0) @safe const nothrow
+    {
+        auto indentStr = ' '.repeat(indent).array;


This will cause an allocation even if trace logging is disabled. Could perhaps turn it into a function?

This will cause an allocation even if trace logging is disabled. Could perhaps turn it into a function?

hmm, Ocean's format function(called by log.trace) is defined as

public void format (Args...) (Level level, cstring fmt, Args args) @safe

so it cannot just lazily evaluate the argument.

the only thing I can think of is

class IndentClass(int indentation) { void toString (scope FormatterSink sink) { sink(' '.repeat(indent).array); } }

and then create an instance of this class on the stack(or just declare it as struct), and then pass that instance to log.trace :)

do you think it is worth the effort?

Maybe it's overkill then. Let's fix it later if it's not a big deal.

There's a simple fix: https://github.com/sociomantic-tsunami/ocean/blob/ba566c473249341633d723904a0b82ba1201a55b/src/ocean/util/log/Logger.d#L385-L391

AndrejMitrovic · 2021-03-15T06:27:59Z

source/agora/consensus/BlockRewardCalc.d

+        Amount gcd_validator_reward = total_validator_reward;
+
+        ulong total_stakes;
+        foreach (ref stake; validator_stakes.byValue())


const ref?

AndrejMitrovic · 2021-03-15T06:29:12Z

source/agora/consensus/BlockRewardCalc.d

+        min_validator_reward.div(total_stakes);
+
+        Amount actual_total_validator_reward;
+        foreach (pkey_stake; validator_stakes.byKeyValue())


Maybe also const ref here?

AndrejMitrovic · 2021-03-15T06:29:45Z

source/agora/consensus/BlockRewardCalc.d

+        remainder.mustSub(actual_total_validator_reward);
+        foreach (amount_point_pair; each_validator_rewards.byKeyValue())
+            log.trace(indentStr ~ "{} -> {}", amount_point_pair.key, amount_point_pair.value);
+        log.trace(indentStr ~ "Remainder is: {}", remainder.toStringNT());


Can use comma instead of concatenation.

AndrejMitrovic · 2021-03-15T06:29:50Z

source/agora/consensus/BlockRewardCalc.d

+        out Amount remainder, double comply_perc, uint indent = 0) @safe const nothrow
+    {
+        auto indentStr = ' '.repeat(indent).array;
+        log.trace(indentStr ~ "Calculating each validator reward");


Can use comma instead of concatenation.

AndrejMitrovic · 2021-03-15T06:34:12Z

source/agora/consensus/data/Params.d

@@ -1,4 +1,4 @@
-/*******************************************************************************
+ 		 /*******************************************************************************


AndrejMitrovic · 2021-03-15T06:35:24Z

source/agora/node/Ledger.d

+        }
+        catch (Exception e) {
+            log.error("unable to read block at height: {}", height);
+            return keys;


Doesn't this just return null then?

ferencdg · 2021-03-15T07:19:24Z

Just a few comments.

I'm not sure whether using floating point is a good idea, there were reservations against using it in the past.

Do you by any chance remember why floating point was not recommended? The only thing I could think is

double is kind of a waste, it takes 64 bytes
some cpu's use different floating point units and precision might differ -> it is not an issue in our case as I think 'double' type is standardized, so arithmetic on them should be the same on every fpu(vs 'real' type in dlang)
precision loss: 0.1 has a finite decimal representation, but infinete binary representation(0.00011001100110011011....) -> we don't care about this

the exponential function that I use for 'block reward reduction' returns a flotaing point value, so the only think I could do is represent the input(comply percentage) to that exponential function in decimal(I just multiply it by 10.000 let's say), so I can pass around that input parameter to different functions. However when I use it as the input in the exponential function, I still have to convert it back to floating point. (I don't think we want to create our own integer based exponential function like 2^x and then try to somehow fit the return value between 0 - 100 so it can be used as reduction percentage)

ferencdg · 2021-03-17T02:04:02Z

thanks for the review @AndrejMitrovic, if anyone else has a comment, please let me know, so this PR could be merged this sprint

omerfirmak · 2021-03-18T07:38:55Z

source/agora/common/Amount.d

+    public string toStringNT () const @safe nothrow
+    {
+        try return this.toString();
+        catch (Exception e) {return "";}


Maybe this return string should indicate the exception.

I was thinking returning null, but that is probably a bad idea: it would require the call site to do extra checks. So as you suggested I will just return some error message.

ferencdg · 2021-03-23T02:37:17Z

rebased, let me know if any more changes needed

AndrejMitrovic · 2021-03-24T05:24:39Z

source/agora/consensus/BlockRewardCalc.d

+       This is because we want to give enough time validators to share
+       block signatures which is the basis of the block reward.
+    3. Each validator will be rewarded based on how many blocks they signed
+       and the reward will NOT be propotional to how much coin they staked.


I don't quite clearly see why it shouldn't be proportional? Is this explained anywhere?

If I remember correctly it was not my decision, but let me know if I am mistaken (@Geod24). A validator who has X * 40k BOA can and will run multiple Agora instances to maximimze its reward, so I agree with you on this.

Geod24

Did a first pass.

Geod24 · 2021-03-23T06:12:10Z

source/agora/common/Set.d

@@ -79,6 +79,12 @@ public struct Set (T)
            this.put(key);
    }

+    /// Returns true if the set has the element e, otherwise returns false
+    public auto opBinary (string op)(in T e) nothrow const pure @safe @nogc if (op == "in")


Type ctor (const, inout, shared, etc...) always come first.
Also it's a template so you don't need to specify all attributes.

not sure I am following this:
'Also it's a template so you don't need to specify all attributes.'

are those attributes stripped after the template instantiation? or are we expecting

return e in _set;

might throw for some instantiation?

I mean that you can remove @safe because the compiler will check the body, see that it is @safe, and mark it @safe itself without you needing to do anything.

Geod24 · 2021-03-23T06:13:02Z

source/agora/common/Amount.d

@@ -86,6 +86,13 @@ public struct Amount
        return ret;
    }

+    /// Non-throwing version of toString
+    public string toStringNT () const @safe nothrow


It's not used anywhere and stands out, I'd remove it.
If you need to debug you can use the scope (failure) assert(0); hack.

Geod24 · 2021-03-23T06:13:37Z

source/agora/test/Common.d

+
+*******************************************************************************/
+
+module agora.test.Common;


That's what Base is for.

'
it's not used anywhere and stands out, I'd remove it.
If you need to debug you can use the scope (failure) assert(0); hack.
'

it is used in BlockRewardCalc.getEachValidatorReward

That's what Base is for.

my plan was to put the higher level classes like Validator/EnrollmentManager to a new file as the module description suggests:

Common classes and utilities shared between the network tests. Classes and utilities in this module are considered higher level than in Base.d.

Base.d is already a monster with 2000+ lines of code. I can move those classes to Base.d though

Why not just make them public in one of the test modules and import them in the other test files. Otherwise common will become the next monster.

I think

It would be hard to find that class if we just put it randomly into some of the test files from where we use it.

Someone already moved those classes into Base.d anyways in recent v0.x.x branch.

In that case can you rebase please.

Geod24 · 2021-03-23T06:14:50Z

source/agora/test/Common.d

+// EnrollmentManager that might or migh not reveal preimage based on the value
+// current value of an atomic shared variable
+version (unittest)
+package class MissingPreImageEM : EnrollmentManager


It's a good idea to make them reusable, but those classes have a few issues. One of them being public static shared UTXOSet utxo_set;

I have an other PR that tries to solve that problem:
#1817 (comment)
any comment on that comment?

I do need to reuse those classes for block reward.

Geod24 · 2021-03-23T06:17:39Z

source/agora/consensus/BlockRewardCalc.d

+
+*******************************************************************************/
+
+module agora.consensus.BlockRewardCalc;


Suggested change

module agora.consensus.BlockRewardCalc;

module agora.consensus.Rewards;

Let's keep it simple and spellable

Geod24 · 2021-03-24T05:18:17Z

source/agora/consensus/BlockRewardCalc.d

+    ///
+    private alias CurrentPeriodTup = Tuple!(ulong, "period", bool, "is_period_end");


A struct would also work :)

Just wondering if struct would have any benefit over Tuples or if we should refine our coding guidelines. Tuples have some nice methods like expand, translate... Any suggestion on when to use one over the other?

Geod24 · 2021-03-24T05:19:37Z

source/agora/consensus/BlockRewardCalc.d

+        min_validator_reward.div(total_stakes);
+
+        Amount actual_total_validator_reward;
+        foreach (const ref pkey_stake; validator_stakes.byKeyValue())


Why the complicated pattern when you can just do:

foreach (key; const ref value; validator_stakes)

?

got a nasty DLang bug when tried:
source/agora/consensus/BlockRewardCalc.d(146,9): Error: _aaApply2 is not nothrow
source/agora/consensus/BlockRewardCalc.d(128,26): Error: nothrow function agora.consensus.BlockRewardCalc.BlockRewardCalc.getEachValidatorReward may throw

to reproduce

void f() nothrow { int[string] m; foreach(const ref k, v; m) { } } int testmain() { f(); return 0; }

Geod24 · 2021-03-24T05:20:01Z

source/agora/consensus/BlockRewardCalc.d

+        auto validator1 = Point.fromString("0xab4f6f6e85b8d0d38f5d5798a4bdc4dd444c8909c8a5389d3bb209a186105110");
+        auto validator2 = Point.fromString("0xab4f6f6e85b8d0d38f5d5798a4bdc4dd444c8909c8a5389d3bb209a186105111");


Use well-known KPs

Geod24 · 2021-03-24T05:23:58Z

source/agora/consensus/BlockRewardCalc.d

+
+    ***************************************************************************/
+
+    public Amount getReducedValidatorReward (in Height height, double comply_perc, uint indent = 0) @safe const nothrow


Let me tell you a floating point joke:
A man has a computer problem. He decides to use floating point. Now he has 2.00000000001 problems.
Usually, we use an integer with a fixed precision for this kind of use case.

ehhmm, I tried to explain it here:
#1768 (comment)

any comment on that?

Geod24 · 2021-03-24T05:24:39Z

source/agora/consensus/BlockRewardCalc.d

+
+        if (reduce_perc == 100)
+            return total_validator_reward;
+        total_validator_reward.div(100);


What about the remainder ?

remainder goes to the Commons Budget and calculated by: getExtraCommonsBudgetReward

ferencdg · 2021-03-25T08:32:34Z

thanks for the review @Geod24, I fixed what I could, but for the following ones I had suggestions/questions:

#1768 (comment)
#1768 (comment)
#1768 (comment)
#1768 (comment)
#1768 (comment)
#1768 (comment)

If I could get some comment on those that you disagree, then I might be able to finish this task this Sprint.

ferencdg · 2021-04-06T04:04:48Z

rebase #3

ferencdg · 2021-05-21T04:08:14Z

rebase #4

hewison-chris

I will continue the review after reading the Rewards section of the white paper.

hewison-chris · 2021-07-06T00:09:42Z

source/agora/consensus/EnrollmentManager.d

+        else
+            return this.keymap.index_to_key[height][index];
+    }
+


Would prefer this in a separate commit (or with the changes it is created for) as it is unrelated to the other changes in this commit.

hewison-chris · 2021-07-06T00:22:16Z

source/agora/test/Common.d

+
+*******************************************************************************/
+
+module agora.test.Common;


Why not just make them public in one of the test modules and import them in the other test files. Otherwise common will become the next monster.

current opDispatch implementaion failes, because 1. return value is void 2. attributes like nothrow are missing

fixes: bosagora#246

ferencdg · 2021-07-15T05:16:45Z

source/agora/node/Ledger.d

@@ -787,11 +929,6 @@ public class Ledger
            return "Internal error: Could not calculate random seed at current height";
        }

-        auto expected_cb_txs = this.getCoinbaseTX(block.txs, block.header.missing_validators);


#1768 (comment)

ferencdg · 2021-07-15T05:17:50Z

rebase #5

hewison-chris · 2021-08-26T08:55:24Z

Closing as will continue in new PR #2415

ferencdg commented Mar 11, 2021

View reviewed changes

ferencdg requested review from AndrejMitrovic and Geod24 March 12, 2021 05:29

ferencdg force-pushed the block-reward branch from 903d997 to 1eab189 Compare March 15, 2021 01:35

AndrejMitrovic reviewed Mar 15, 2021

View reviewed changes

ferencdg force-pushed the block-reward branch 2 times, most recently from 13fa65a to f7ca780 Compare March 17, 2021 01:59

omerfirmak reviewed Mar 18, 2021

View reviewed changes

ferencdg force-pushed the block-reward branch 2 times, most recently from f7875c4 to e27b616 Compare March 23, 2021 02:36

AndrejMitrovic reviewed Mar 24, 2021

View reviewed changes

Geod24 reviewed Mar 24, 2021

View reviewed changes

ferencdg mentioned this pull request Mar 24, 2021

mustAdd/mustSub shouldn't be used outside of tests #1824

Closed

ferencdg force-pushed the block-reward branch 2 times, most recently from 0957ff0 to 570c912 Compare March 25, 2021 08:13

ferencdg force-pushed the block-reward branch from 570c912 to 5c69fe3 Compare March 25, 2021 08:46

ferencdg mentioned this pull request Mar 26, 2021

Modified style guide: func type ctr preceds attr #1827

Merged

ferencdg force-pushed the block-reward branch from 5c69fe3 to 8f4a302 Compare April 6, 2021 04:05

ferencdg force-pushed the block-reward branch 2 times, most recently from ed5f4e2 to e2a775e Compare May 20, 2021 08:01

ferencdg force-pushed the block-reward branch from e2a775e to c17f300 Compare May 21, 2021 03:30

ferencdg force-pushed the block-reward branch from c17f300 to da52927 Compare May 21, 2021 05:59

hewison-chris reviewed Jul 6, 2021

View reviewed changes

Daniel Graczer added 6 commits July 14, 2021 14:22

Added 'enabled' method to Log.d

746766c

current opDispatch implementaion failes, because 1. return value is void 2. attributes like nothrow are missing

Amending maximum coin supply based on whitepaper

7654bf2

Overrode operator 'in' method in Set

91b2b5f

Added attributes to ConsensusParams' ctr

48cecb0

Added non-throwing version of toString

9821469

Added block reward

3055b73

fixes: bosagora#246

ferencdg force-pushed the block-reward branch from da52927 to 3055b73 Compare July 15, 2021 05:14

ferencdg commented Jul 15, 2021

View reviewed changes

hewison-chris closed this Aug 26, 2021

		@@ -0,0 +1,87 @@
		/*******************************************************************************

		@@ -1,4 +1,4 @@
		/*******************************************************************************
		/*******************************************************************************


		*******************************************************************************/

		module agora.test.Common;


		*******************************************************************************/

		module agora.consensus.BlockRewardCalc;

	module agora.consensus.BlockRewardCalc;
	module agora.consensus.Rewards;

		///
		private alias CurrentPeriodTup = Tuple!(ulong, "period", bool, "is_period_end");

		auto validator1 = Point.fromString("0xab4f6f6e85b8d0d38f5d5798a4bdc4dd444c8909c8a5389d3bb209a186105110");
		auto validator2 = Point.fromString("0xab4f6f6e85b8d0d38f5d5798a4bdc4dd444c8909c8a5389d3bb209a186105111");


		***************************************************************************/

		public Amount getReducedValidatorReward (in Height height, double comply_perc, uint indent = 0) @safe const nothrow

Define and implement block rewards #1768

Define and implement block rewards #1768

Conversation

ferencdg commented Mar 11, 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

ferencdg Mar 11, 2021 • edited

Choose a reason for hiding this comment

ferencdg Mar 11, 2021 • edited

Choose a reason for hiding this comment

codecov bot commented Mar 11, 2021 • edited

Codecov Report

ferencdg commented Mar 15, 2021

AndrejMitrovic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ferencdg Mar 15, 2021 • edited

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

ferencdg Mar 17, 2021 • edited

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

ferencdg commented Mar 15, 2021 • edited

ferencdg commented Mar 17, 2021

Choose a reason for hiding this comment

ferencdg Mar 18, 2021 • edited

Choose a reason for hiding this comment

ferencdg commented Mar 23, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Geod24 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

Choose a reason for hiding this comment

ferencdg Mar 24, 2021 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ferencdg Jul 6, 2021 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ferencdg Mar 24, 2021 • edited

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

ferencdg commented Mar 25, 2021 • edited

ferencdg commented Apr 6, 2021

ferencdg commented May 21, 2021

hewison-chris 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

ferencdg commented Jul 15, 2021

hewison-chris commented Aug 26, 2021

ferencdg Mar 11, 2021 •

edited

ferencdg Mar 11, 2021 •

edited

codecov bot commented Mar 11, 2021 •

edited

ferencdg Mar 15, 2021 •

edited

ferencdg Mar 17, 2021 •

edited

ferencdg commented Mar 15, 2021 •

edited

ferencdg Mar 18, 2021 •

edited

ferencdg Mar 24, 2021 •

edited

ferencdg Jul 6, 2021 •

edited

ferencdg Mar 24, 2021 •

edited

ferencdg commented Mar 25, 2021 •

edited