ESIP-8: Ethscription Attachments aka "BlobScriptions"

Links

• Reference Implementation

• ESIP-8 Discussion

Abstract

The introduction of blobs in EIP-4844 enables anyone to store data on Ethereum for 10x to 100x cheaper than calldata. This comes at a cost, however: the Ethereum protocol doesn't guarantee the availability of blob data for more than 18 days.

However, on a practical level it is not clear how burdensome this limitation will be. Because L2s use blobs to store transaction data there will be strong incentives to create publicly accessible archives of blob data to enhance the transparency and auditability of Layer 2s.

Also, like IPFS, blob data is completely decentralized—as long as one person has blob data it can be verified and used by anyone.

This ESIP proposes using blobs to store data within the Ethscriptions Protocol. We presuppose the ready availability of blob data and require indexers to store or find user blob data along with the other blockchain data the Ethscriptions Protocol currently uses.

Specifically, ESIP-8 proposes a new "sidecar" attachment field for Ethscriptions that is composed from the data in one or more blobs. This field is in addition to the existing content field.

The name "Ethscription Attachment" is preferred over "Ethscription Blob" (or similar) because transactions can have multiple blobs, but ethscriptions can only have one attachment (that is composed of all the blobs together).

An Example

Consider the ethscription created by this Sepolia transaction. The transaction's calldata contains the hex data 0x646174613a2c68656c6c6f2066726f6d20457468736372697074696f6e2063616c6c6461746121 which corresponds to the dataURI "data:,hello from Ethscription calldata!" which becomes the ethscription's content.

The transaction's blobs, when interpreted according to the rules described below, contains the data for this image which becomes the ethscription's attachment:

Specification

All new ethscriptions have an optional attachment field. If an ethscription is created in a transaction with no blobs this field will be null.

If an ethscription's creation transaction does include blobs and the ethscription was created via calldata (i.e., not via an event emission), its blobs are concatenated and interpreted as an untagged CBOR object (as defined by RFC 8949) that decodes into a hash with exactly these keys:

• content

• contentType

If the concatenated data is a valid CBOR object, and that object decodes into a hash with exactly those two fields, an attachment for the ethscription is created.

The case in which the blobs are invalid and an attachment is not created is handled identically to the case in which there are no blobs at all. I.e., the ethscription is still created if it's otherwise valid, just with no attachment.

Note:

• There is no uniqueness requirement for the attachment's content and/or contentType.

• Attachment content, contentType, and the container CBOR object itself can each be optionally gzipped with a maximum compression ratio of 10x.

• The attachment is not valid if:

  • If the CBOR object has a tag

  • If the decoded object his not a hash

  • If the decoded hash's keys aren't exactly content and contentType. There cannot be extra keys.

  • The values of content and contentType aren't both strings (either binary or UTF-8).

When such an attachment exists, the indexer's API must include the path for retrieving it in an attachment_path field in the JSON representation of an ethscription with at most a one block delay between ethscription creation and inclusion of the URL. For example, if an ethscription is created in block 15, the attachment_url must appear no later than block 17.

The attachment_url field will be available in addition to the content_uri field.

contentType Max Length

To enable performant filtering by contentType, indexers must only store the first 1,000 characters of the user-submitted content type. Content types of more than 1,000 characters will be truncated, but the attachment will still be valid.

Creating an Ethscription Attachment in Javascript

You can use the cbor package and Viem's toBlobs:

const { toBlobs } = require('viem');
const fs = require('fs');
const cbor = require('cbor');

const imagePath = '/whatever.gif'
const imageData = fs.readFileSync(imagePath);

const dataObject = {
  contentType: 'image/gif',
  content: imageData
};

const cborData = cbor.encode(dataObject);
const blobs = toBlobs({ data: cborData });

Getting Blob Data

Blob data is available on a block-level through the blob_sidecars API endpoint available on Ethereum Beacon nodes. If you don't want to run a node yourself, you can use Quicknode.

The input to this function is a "block id," which is a slot number (not block number) or block root. Block roots are available on normal Ethereum API requests, but only for the previous block (the field is parentBeaconBlockRoot).

This means that attachments must be populated on a one block delay.

Associating Blobs with Ethereum Transactions

Because the Beacon API only provides blob information on a block level, it requires some additional logic to match blobs to the transactions that created them. Fortunately, transactions now have a blobVersionedHashes field that can be computed from the kzg_commitments field on the block-level blob data.

Here's an example implementation (it's O(n^2) but the numbers involved are small)

def transaction_blobs
  blob_versioned_hashes.map do |version_hash|
    blob_from_version_hash(version_hash)
  end
end

def blob_from_version_hash(version_hash)
  block_blob_sidecars.find do |blob|
    kzg_commitment = blob["kzg_commitment"].sub(/\A0x/, '')
    binary_kzg_commitment = [kzg_commitment].pack("H*")
    sha256_hash = Digest::SHA256.hexdigest(binary_kzg_commitment)
    modified_hash = "0x01" + sha256_hash[2..-1]
    
    version_hash == modified_hash
  end
end

Converting Blob Content to an Attachment

At a high-level we use normalize the blob data, concatenate it, and CBOR-decode it. However blobs have a few interesting quirks that make this more challenging:

• Currently blobs have a minimum length of 128kb. If your data is smaller than that you'll have to pad it (probably will null bytes) to the full length.

• Blobs are composed of "segments" of 32 bytes, none of which, when interpreted as an integer, can exceed the value of the cryptography-related "BLS modulus", which is 52435875175126190479447740508185965837690552500527637822603658699938581184513.

So if you want to use blobs you need a protocol for communicating where the data ends and a mechanism for ensuring no 32 byte segment is too large.

Here Ethscriptions will follow Viem's approach:

• Left-pad each segment with a null byte. A 0x00 in the most significant byte ensures no segment can be larger than the BLS modulus.

• End the content of every blob with 0x80, which, when combined with the rule above, provides an unambiguous way to determine the length of the data in the blob.

When a blob creator follows these rules (or just use's Viem's toBlobs), you can decode it into bytes by using Viem's fromBlobs. There is a Ruby implementation as well in the appendix.

Once you decode the blob you can create an attachment using something like this class.

class EthscriptionAttachment < ApplicationRecord
  class InvalidInputError < StandardError; end
  
  has_many :ethscriptions,
    foreign_key: :attachment_sha,
    primary_key: :sha,
    inverse_of: :attachment
  
  delegate :ungzip_if_necessary!, to: :class
  attr_accessor :decoded_data
  
  def self.from_eth_transaction(tx)
    blobs = tx.blobs.map{|i| i['blob']}
    
    cbor = BlobUtils.from_blobs(blobs: blobs)

    from_cbor(cbor)
  end
  
  def self.from_cbor(cbor_encoded_data)
    cbor_encoded_data = ungzip_if_necessary!(cbor_encoded_data)
    
    decoded_data = CBOR.decode(cbor_encoded_data)
    
    new(decoded_data: decoded_data)
  rescue EOFError, *cbor_errors => e
    raise InvalidInputError, "Failed to decode CBOR: #{e.message}"
  end
  
  def decoded_data=(new_decoded_data)
    @decoded_data = new_decoded_data
    
    validate_input!
    
    self.content = ungzip_if_necessary!(decoded_data['content'])
    self.content_type = ungzip_if_necessary!(decoded_data['contentType'])
    self.size = content.bytesize
    self.sha = calculate_sha
    
    decoded_data
  end
  
  def calculate_sha
    combined = [
      Digest::SHA256.hexdigest(content_type),
      Digest::SHA256.hexdigest(content),
    ].join
    
    "0x" + Digest::SHA256.hexdigest(combined)
  end
  
  def self.ungzip_if_necessary!(binary)
    HexDataProcessor.ungzip_if_necessary(binary)
  rescue Zlib::Error, CompressionLimitExceededError => e
    raise InvalidInputError, "Failed to decompress content: #{e.message}"
  end
  
  private
  
  def validate_input!
    unless decoded_data.is_a?(Hash)
      raise InvalidInputError, "Expected data to be a hash, got #{decoded_data.class} instead."
    end
    
    unless decoded_data.keys.to_set == ['content', 'contentType'].to_set
      raise InvalidInputError, "Expected keys to be 'content' and 'contentType', got #{decoded_data.keys} instead."
    end
    
    unless decoded_data.values.all?{|i| i.is_a?(String)}
      raise InvalidInputError, "Invalid value type: #{decoded_data.values.map(&:class).join(', ')}"
    end
  end
  
  def self.cbor_errors
    [CBOR::MalformedFormatError, CBOR::UnpackError, CBOR::StackError, CBOR::TypeError]
  end
end

Hashing Attachments

It's useful to be able to generate a unique hash of an Ethscription Attachment in order for indexers to avoid storing duplicate data and for users to determine which other ethscriptions have the same attachment. This can be done in many ways, but to promote uniformity ESIP-8 defines this canonical method of hashing Ethscription Attachments:

1. Compute the sha256 hash of the attachment's ungzipped contentType and content fields.

2. Remove the leading 0x if present.

3. Concatenate the hex string representations of the hashes with the contentType hash first.

4. Hash this concatenated string and add a 0x prefix.

Here is a Javascript implementation:

import { sha256, stringToBytes } from 'viem';

const attachment = {
  contentType: 'text/plain',
  content: 'hi',
};

const contentTypeHash = sha256(stringToBytes(attachment.contentType));
const contentHash = sha256(stringToBytes(attachment.content));

const combinedHash =
  contentTypeHash.replace(/^0x/, '') + contentHash.replace(/^0x/, '');

const finalHash = sha256(combinedHash as `0x${string}`);

And a Ruby implementation:

require 'digest'

attachment = {
  'contentType' => 'text/plain',
  'content' => 'hi',
}

content_type_hash = Digest::SHA256.hexdigest(attachment['contentType'])
content_hash = Digest::SHA256.hexdigest(attachment['content'])

combined_hash = content_type_hash + content_hash

final_hash = "0x" + Digest::SHA256.hexdigest(combined_hash)

Appendix: Ruby BlobUtils

module BlobUtils
  # Constants from Viem
  BLOBS_PER_TRANSACTION = 2
  BYTES_PER_FIELD_ELEMENT = 32
  FIELD_ELEMENTS_PER_BLOB = 4096
  BYTES_PER_BLOB = BYTES_PER_FIELD_ELEMENT * FIELD_ELEMENTS_PER_BLOB
  MAX_BYTES_PER_TRANSACTION = BYTES_PER_BLOB * BLOBS_PER_TRANSACTION - 1 - (1 * FIELD_ELEMENTS_PER_BLOB * BLOBS_PER_TRANSACTION)

  # Error Classes
  class BlobSizeTooLargeError < StandardError; end
  class EmptyBlobError < StandardError; end
  class IncorrectBlobEncoding < StandardError; end

  # Adapted from Viem
  def self.to_blobs(data:)
    raise EmptyBlobError if data.empty?
    raise BlobSizeTooLargeError if data.bytesize > MAX_BYTES_PER_TRANSACTION
    
    if data =~ /\A0x([a-f0-9]{2})+\z/i
      data = [data].pack('H*')
    end

    blobs = []
    position = 0
    active = true

    while active && blobs.size < BLOBS_PER_TRANSACTION
      blob = []
      size = 0

      while size < FIELD_ELEMENTS_PER_BLOB
        bytes = data.byteslice(position, BYTES_PER_FIELD_ELEMENT - 1)

        # Push a zero byte so the field element doesn't overflow
        blob.push(0x00)

        # Push the current segment of data bytes
        blob.concat(bytes.bytes) unless bytes.nil?

        # If the current segment of data bytes is less than 31 bytes,
        # stop processing and push a terminator byte to indicate the end of the blob
        if bytes.nil? || bytes.bytesize < (BYTES_PER_FIELD_ELEMENT - 1)
          blob.push(0x80)
          active = false
          break
        end

        size += 1
        position += (BYTES_PER_FIELD_ELEMENT - 1)
      end

      blob.fill(0x00, blob.size...BYTES_PER_BLOB)
      
      blobs.push(blob.pack('C*').unpack1("H*"))
    end

    blobs
  end
  
  def self.from_blobs(blobs:)
    concatenated_hex = blobs.map do |blob|
      hex_blob = blob.sub(/\A0x/, '')
      
      sections = hex_blob.scan(/.{64}/m)
      
      last_non_empty_section_index = sections.rindex { |section| section != '00' * 32 }
      non_empty_sections = sections.take(last_non_empty_section_index + 1)
      
      last_non_empty_section = non_empty_sections.last
      
      if last_non_empty_section == "0080" + "00" * 30
        non_empty_sections.pop
      else
        last_non_empty_section.gsub!(/80(00)*\z/, '')
      end
      
      non_empty_sections = non_empty_sections.map do |section|
        unless section.start_with?('00')
          raise IncorrectBlobEncoding, "Expected the first byte to be zero"
        end
        
        section.delete_prefix("00")
      end
      
      non_empty_sections.join
    end.join
    
    [concatenated_hex].pack("H*")
  end
end