A diamond implementation implements EIP-2535 Diamonds. These are diamond reference implementations.

Diamonds enable you to build efficient, powerful, flexible, modular contract systems.

Every diamond implementation implements the following:

1. diamondCut function Standard function used to add/replace/remove functions on a diamond.
2. Loupe functions Four standard functions used to show what functions and facets a diamond has.

There are three diamond reference/example implementations that have different benefits and disadvantages in terms of code complexity and gas costs. Here is a breakdown of the differences between the three implementations. The ratings (high, medium, low) are in relation to each other.

It is possible to choose one implementation and then in the future upgrade the diamond to switch to a different implementation.

All three implementations pass the same tests.

diamond-1 and diamond-2 use less gas to add/replace/remove functions.

diamond-3 uses less gas to call the diamond loupe functions.

diamond-1 and diamond-2 are implemented the same way except that diamond-2 is gas optimized. To understand how diamond-2 is implemented look at diamond-1 first.

Links to diamond reference implementation repositories:

• diamond-1-hardhat
• diamond-2-hardhat
• diamond-3-hardhat

1. Has an bytes4[] selectors array that stores the function selectors of a diamond.
2. Has a mapping(bytes4 => FacetAddressAndSelectorPosition) facetAddressAndSelectorPosition mapping that maps each function selector to its facet address and its position in the selectors array.

It's facets, facetFunctionSelectors, facetAddresses loupe functions should not be called in on-chain transactions because their gas cost is too high. These functions should be called by off-chain software.

The facetAddress loupe function has a low fixed gas cost in all implementations and can be called in on-chain transactions.

diamond-2 is implemented the same way as diamond-1 except that the selectors array is implemented as a mapping of 32-byte storage slots and uses various gas-optimizations to reduce storage reads and writes.

This implementation is gas-optimized for adding/replacing/removing functions on a diamond.

It's facets, facetFunctionSelectors, facetAddresses loupe functions should not be called in on-chain transactions because their gas cost is too high. These functions should be called by off-chain software.

The facetAddress loupe function has a low gas cost in all implementations and can be called in on-chain transactions.

1. Has an address[] facetAddresses array that stores the facet addresses of a diamond.

2. Has a mapping(address => FacetFunctionSelectors) facetFunctionSelectors mapping that maps each facet address to its array of function selectors and its position in the facetAddresses array.

3. Has a mapping(bytes4 => FacetAddressAndPosition) selectorToFacetAndPosition mapping that maps each function selector to its facet address and its position in the facetFunctionSelectors[facetAddress].functionSelectors array.

The standard loupe functions facets, facetFunctionSelectors, facetAddresses CAN be called in on-chain transactions. Note that if a diamond has a great many functions and/or facets these functions may still cause an out-of-gas error.

The facetAddress loupe function has a low fixed gas cost in all implementations and can be called in on-chain transactions.