Gas refunds are provided when clearing storage or calling SELFDESTRUCT on contracts.

The Yellow Paper mentions that refunds are "being capped up to a maximum of half...". What exactly is this "half" and the other components that sum up to a refund? Preferred answer will explain with a specific example (since the Yellow Paper has the definitions).

What are the limits? What attacks would have been possible if refunds were not limited?

  • Does shortening an array do this type of SSTORE 0? Feb 21 '18 at 0:27
  • @FullDecent No SSTORE done when setting an array's length (I checked Remix).
    – eth
    Feb 28 '18 at 6:34

The following opcodes trigger a refund

  • SSTORE[x] = 0 (i.e. deletion)

SELFDESTRUCT refunds 24.000 and SSTORE clear refunds 15.000 gas. The accumulated refund can not exceed half the gas used for the current context (i.e. the initial call).

Let's take the following example:

Current state of the contract's storage

0x00: 1
0x01: 1

And the following execution

SSTORE ; set contents of address 0x0 to 0 (i.e. delete)

This would result in the following gas sum 21000 + 3 + 3 + 5000 (tx_gas + push_gas + push_gas + sstore_clear_gas). The total amount of which may be refunded is therefor gas_sum / 2 (13.003). The amount refunded is 15.000 but since this clearly exceeds the maximum of 13.003 we cap it to that amount.

And the reason why we cap this is so that the miner does not end up paying for the actual execution of the contract :-)


If the refunds weren't capped we would run in to 2 issues:

  1. It disincentives running any contracts that refund;
  2. Miners may end up paying for the execution.

It would be disincentive to run these type of transactions if they weren't capped because at the end of the execution the refund is reduced from the amount that which a miner gets (i.e. gasUsed * tx.price). Capping at say the gasUsed would render the execution useless to the miner (it doesn't get rewarded for running the contract). This brings us to the second point; having a higher refund than the gas used would therefor end up in a negative gas usage and would therefor have to be reduced from the miner's balance. No miner in their right mind is going to pay you for execution your call.

  • with 1) you mean "it disincentives miners running any contracts that refund" right? not a 'user'? Jun 21 '16 at 10:30
  • Indeed @RubendeVries
    – Jeffrey W.
    Jun 21 '16 at 12:26
  • 1
    In other words, if I understand correctly, you don't want people to create a bunch of contracts during quiet times with low gas price and then suicide() them during an ICO with high gas prices. Aug 22 '17 at 16:58
  • This is a great explanation and it'd be really useful to have it in the yellow paper! Jan 5 '19 at 19:06
  • @JeffreyW. wrong Answer! Used gas and remaining gas are 2 differents things. Used gas increase each times an Instrcution is executed, thereby (and unlike some explorers are suggesting), refunding doesn t decrease miner rewards. But it s worse than that: the person who launched the transaction would still get more ether than what was paid in gas: they would be genereted out of nothing resulting in inflation! This is clearer when you read the source code from officials implementations than the yellow paper. Jul 24 '19 at 11:24

Another important "limit" is that the gas refund is provided only 1 time, at the very end of the transaction's execution.

This means that a transaction cannot use its own gas refund to pay for any part of its execution.

A refund cannot be used to avoid Out of Gas exceptions, as illustrated by these examples:

How come I need to specify a higher gasLimit than estimateGas?

How to clear large arrays without blowing the gas limit?

  • Indeed. This would destroy the security of .send and .transfer if gas refund was provided as it was burned, as the receiving contract could use its gas stores to reenter into the calling function and change its state. Nov 16 '18 at 0:18

The reason for 50% gas refund cap is given in Design Rationale:

The 50% cap is needed to ensure that a miner given a transaction with some quantity of gas can still determine an upper bound on the computational time to execute the transaction.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.