Option to not store Swarm hash of a metadata file into the bytecode

[ethernomad]

It isn't possible to use the etherscan.io verification facility now that this functionality has been added.

Can it be made a command line option?

[chriseth]

Can you explain a bit more why this breaks etherscan verification? The idea was that it actually helps verification...

[ethernomad]

The problem is that the bytecode is now dependent on more things. You can use the same compiler version, but if you use Linux instead of browser solidity or Windows you will get a different hash.

Another problem is the import directive. etherscan doesn't support this as you have to paste in a single file. Normally you would get around this problem by manually replacing the import directive with the contents of the imported file. But now this changes the hash and it won't verify.

[chriseth]

Do you have examples of getting different hashes on linux and windows? I assume this is mostly related to file names because they are part of the metadata and thus of the bytecode.

I would not recommend replacing the import directive by the file contents, this modifies the code. Instead, please call the compile with multiple inputs.

So my feeling is that requiring the user to supply file names or even better, require them to publish the metadata file instead of just the source code, would solve the problem.

[ethernomad]

The operating system is in the metadata:

{"compiler":{"version":"0.4.8+commit.60cc1668.Linux.g++"}

Replacing import directives with file contents is a requirement of how etherscan's verification works.

I'm sure a better verification system can be built using the metadata hash. My request is that there should still be the option to use the existing etherscan verification system with the latest compiler until etherscan deploy a new system that uses the metadata hash.

[chriseth]

If etherscan modifies the source in such a way, then things like import {X} from "b.sol"; or import "b.sol" as Y; do not work, and I guess that the larger contract systems make heavy use of this.

I really want to urge people to use the metadata hash, so I don't think that removing it is a good way to push it forward. Would it be an option to ignore the metadata hash in the etherscan's verification procedure and also warn the users about this?

[axic]

While I like having these optional, the goal is to have it in a way which wouldn't interfere with other ways.

@mtbitcoin As an intermediate measure, would it be possible to add the hash parsing to etherscan and ignore that bit when comparing the outputs? See this code how to match the hash: https://github.com/ethereum/browser-solidity/blob/master/src/app/renderer.js#L94

And here's a more detailed explanation: http://solidity.readthedocs.io/en/develop/miscellaneous.html?highlight=metadata#encoding-of-the-metadata-hash-in-the-bytecode

[ethernomad]

Okay I guess it's up to etherscan to fix their verification system.

[axic]

As per future verification work, having different versions could manifest as problems, albeit not necessarily:

{"compiler":{"version":"0.4.8+commit.60cc1668.Linux.g++"}

We aim to use this to lookup the compiler version, and very likely, most implementations would resort to using solc-js to rerun the compilation.

For verification, the bytecode would need to be compared entirely, including the hash. This will obviously fail if it was compiled with a Linux or Windows solc.

Either we ignore the hash or truncate the version number not to include the OS.

[chriseth]

Note that the behaviour of some weird contracts might depend on the metadata hash as you can read it using assembly, so a proper verification also has to check the metadata hash.

[holiman]

Wouldn't a solution be to have an option to the compiler, --metadata='0.4.8+commit.60cc1668.Linux.g++'. So if you want to verify, just take the metadata from the contract?

[chriseth]

@axic @holiman yes, I think we should just truncate the version. If it does not compile the same way on a different operating system the security of the contract is in danger anyway. I also considered @holiman's suggestion of adding a "fake version string", but I think that opens the door for manipulation.

[chriseth]

If you really rely on some specific operating system / architecture, you can provide that as a comment in the source code.

[axic]

The output of the compiler on all platforms should match and if it doesn't we have a bug to fix 😉, therefore truncating the version should be safe.

[axic]

Wouldn't a solution be to have an option to the compiler, --metadata='0.4.8+commit.60cc1668.Linux.g++'

I'm strongly against adding an option to modify the version number from the outside.

[holiman]

but I think that opens the door for manipulation.

Hold on - we can't assume that this wont be manipulated, and shouldn't place any trust in that piece of data. It's just for aiding verification, but if it's going to be used for 'cheating' by skipping verification altogether, well, that sounds like a bad idea.