How does the EVM identify a dynamic type in the function calldata?

Question

So as per the ABI encoding specifications mentioned in Solidity Docs, for a dynamic array, the data location (in bytes from start, for the data part) is specified at first, followed by the length of the array at the previously specified location, followed by the actual data values.

Example in docs:

Function to be called: function sam(bytes memory, bool, uint[] memory) public pure {}

Calldata bytes for sam("dave", true, [1, 2, 3]):

0xa5643bf20000000000000000000000000000000000000000000000000000000000000060000000000000000000000000000000000000000000000000000000000000000100000000000000000000000000000000000000000000000000000000000000a0000000000000000000000000000000000000000000000000000000000000000464617665000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000003000000000000000000000000000000000000000000000000000000000000000100000000000000000000000000000000000000000000000000000000000000020000000000000000000000000000000000000000000000000000000000000003

Here, the first set of bold bytes represents the location of the bytes value for "dave". The second set of bold values is the actual data (starting with the length of the data which is 4 here).

So, while interpreting the calldata, how does the EVM differ the location of the dynamic data from being just a normal variable value? i.e here, how is 0x00...60 is identified as a location value, instead of being maybe a static uint?

Answer 1

Answering my own question.

So, I apparently missed the fact that the abi-encoding function follows the head-tail encoding mechanism. The abi-decoding function then keeps track of the parameter type- actually it checks if the data type has a predefined size, if not, then it treats the value in hand as a start position.

Here is a very clear pyethereum implementation of decode_abi.

This is the magical part of the function:

for i, typ in enumerate(types):
    if sizes[i] is None:
        start_positions[i] = big_endian_to_int(data[pos:pos + 32])
        j = i - 1
        while j >= 0 and start_positions[j] is None:
            start_positions[j] = start_positions[i]
            j -= 1
        pos += 32
    else:
        outs[i] = data[pos:pos + sizes[i]]
        pos += sizes[i]