Not supported evm.bytecode and evm.assembly options in compiler configuration

[smiasojed]

Although the tool claims that it does not support evm.bytecode and evm.assembly, it still returns it as part of the compilation output.

Used compiler config:

{
	"language": "Solidity",
	"settings": {
		"optimizer": {
			"enabled": true,
			"runs": 200
		},
		"outputSelection": {
			"*": {
			"": ["ast"],
			"*": ["abi", "metadata", "devdoc", "userdoc", "storageLayout", "evm.bytecode", "evm.legacyAssembly", "evm.deployedBytecode", "evm.methodIdentifiers", "evm.assembly"]
			}
		}
	}
}

It is expected that tool accepts evm.bytecode and evm.assembly

[smiasojed]

evm.gasEstimates is also not supported

[xermicus]

Stemming from an internal discussion:

• Provide what was requested; however
• Provide default values for incompatible data (if requested)

We substitute any byte-code incompatible fields with empty objects. This won't crash any tools relying on it, they can still request it but will get a zero or empty value. Which they can do nothing with so we should be save. Also this will make remix happy.

[athei]

Both of them should be advertised as being available in order to maximize compatibility.

evm.bytecode should be returned but empty (we only have init code). evm.assembly should contain the PolkaVM dissassembly.

[xermicus]

To me it'd also make sense to just provide the blob for both, the deploy code and runtime code? Because it is the same, it seems like we can just provide both for maximum compatibility.

[athei]

Yeah that would work, too.

[smiasojed]

Remix uses evm.bytecode internally, so I think it is better to keep bytecode in sync with deployedBytecode.

[athei]

Because of course they do. They depend on every little detail :(

[smiasojed]

evm.assembly should contain the PolkaVM dissassembly.

I think that we need to update polkavm-disassembler to be compatible with evm.legacyAssembly

"legacyAssembly": 
				{
							".code": [
								{
									"begin": 199,
									"end": 555,
									"name": "PUSH",
									"source": 0,
									"value": "80"
								}
                                                           ]
				}

[xermicus]

Legacy assembly is not to confuse with EVM bytecode. Legacy assembly is technically speaking on of our IRs. It can be passed through as-is.

[smiasojed]

We have in revive

  #[serde(rename = "legacyAssembly")]
    pub assembly: Option<Assembly>,
    /// The contract PolkaVM assembly code.
    #[serde(rename = "assembly")]
    pub assembly_text: Option<String>,

assembly_text (evm.assembly) is updated with polkavm assembly but assembly (evm.legacyAssembly) is still evm assembly

[xermicus]

Ah yeah, assembly_text is custom, it isn't a thing for solc.

[xermicus]

I reckon either assembly and legacyAssembly could as well be removed (i.e. replaced with the empty value). We will have maximum compat by keeping it. However I don't see what tools are supposed to do with this.

[smiasojed]

evm.assembly contains assembly_text, so I think it is good to keep it.
However, Remix displays evm.legacyAssembly, which is incorrect.
The options are to either include PolkaVM assembly there or leave it empty.

[xermicus]

Ah okay, I see where you are coming from. The problem is we are adding one more IR format. Remix is selling evm.legacAssembly as the "assembly" code. Which one would assume to be PVM assembly in our case. If we substitute it, we introduce some amount of confusion. But we have to do this anyways for the bytecode, in order to make toolings "just work". For me it's fine to substitute.

[smiasojed]

Remix parses this before displaying and expect JSON format. We can not just substitute it, because parsing will fail.
In my opinion we have 2 options:

1. empty object
2. polkavm assembly in right format (Requires a change in polkavm-disassembler to write raw instructions)

[xermicus]

Why would 2. require a change in polkavm-disassembler instead bringing it to the right format directly in revive? Tying the PVM disassembler to this seems a bit stretched 😅

[xermicus]

In case the formats are inherently incompatible, we should change our remix instance to display what it finds in assembly_text.

[xermicus]

Stemming from an internal discussion:

* Provide what was requested; however

* Provide default values for incompatible data (if requested)

We substitute any byte-code incompatible fields with empty objects. This won't crash any tools relying on it, they can still request it but will get a zero or empty value. Which they can do nothing with so we should be save. Also this will make remix happy.

So if they are incompatible, seems like we hit the second case here. Replacing it with empty values and instead use assembly_text in our remix fork seems the way to go. Then we don't break any tools but our remix instance can still display it.

[smiasojed]

Why would 2. require a change in polkavm-disassembler instead bringing it to the right format directly in revive? Tying the PVM disassembler to this seems a bit stretched 😅

Now looking into it maybe we could indeed parse the polkavm assembly and output something similar to legacyAssembly,
but I am not sure if it make sense.

Legacy assembly is built from objects containing
Name: The opcode (e.g., ADD).
in polkavm we get s0 = a0 + a1 so we need to parse the instructions and replace it with ADD
Value: The operand or value pushed to the stack.
In our case we could have registers [s0, a0, a1]
Source: The source code location (related to Solidity source code).
I think, we do not have it
Begin and End: The bytecode offsets where the instruction starts and ends.
We could get it from disassembler

[xermicus]

I vote for just using assembly_text directly in remix.

[xermicus]

@smiasojed is this issue still relevant?

[smiasojed]

@smiasojed is this issue still relevant?

yes, I need to do still something with evm.gasEstimates. I will handle it when I have some time.