call causes exponential gas explosion about klab

[livnev]

I'm concerned that these can't be algebraically simplified because of the nonlinear / 64

[mhhf]

I disagree, see the structure of it and my proposed rule where:

A := (VGas - 17080)
B := 700
C := ((VGas - 17780) - ((VGas - 17780) / 64))
D := 636

the simplified version after the rule applied would be:

((VGas - 17080) - 700) - 636 

which would further simplify

similar stuff can be done to nested #if things.

mr. martin noted tho that we would get problems with the gas analyzer due to this lemma since the gas behaves non-monotonic. But We can recover this information (C) in a different way when we scan all #refund rules.
we need to spend some time on all this :/

[livnev]

You're right, I didn't see that there was actually repetition there, so the expression can be simplified.

@MrChico is right that the gas analyser works by detecting these / 64 subexpressions and making sure they are all >= 0 to ensure there is no OOG inside of a call.

What if we keep this stuff unsimplified during a run but have the gas analyser apply this simplification to the post-gas, after it has already used the unsimplified / 64 subexpressions for solving?

How do you want to keep track of the #refund rules? I think we discussed once adding a cell to evm-semantics that tracks a list of ending gas expressions in every message call: that would be very useful here...

[mhhf]

What if we keep this stuff unsimplified during a run but have the gas analyser apply this simplification to the post-gas, after it has already used the unsimplified / 64 subexpressions for solving?

the whole point of this issue is to supply k with lemmas which reduce the gas DURING a run not after.

the #refund rules could be marked and gathered in the same way we do it with evm steps.
Then depending on what we need we would get either a list of them or a branching tree.

[livnev]

@mhhf OK, then what we will need is every time the #refund rule is applied, with <gas> GAVAIL </gas>, save the GAVAIL into a list.

[mhhf]

@livnev how would you like to receive this?
is an array of json objects containing the arithmetic expression ok?

[livnev]

@mhhf Yes. And you will keep them in json form right, without printing them?

[mhhf]

yep, raw json
looks like the pr is getting merged soon
i'll update klab and evm-semantics right after so you should be able to play around with it on thursday

[livnev]

I believe we have resolved this with the new lemmas plus the improvements to the solver.