will on Warpcast

Content pfp

https://warpcast.com/~/channel/ponder-sh

0 reply

0 recast

0 reaction

Kevin pfp

@typedarray.eth

This remains unsolved. I think it's impossible - if anyone wants to learn more about traces and double check my work, would gladly chat

3 replies

1 recast

25 reactions

Kevin pfp

@typedarray.eth

Created a gist with examples + more context https://gist.github.com/0xOlias/596f21266023bc79adcff3f4d2c113b7 TL;DR - our requirements are weird, but if we solve this it will save Ponder users lots of time and money (on their RPC bill 🙃)

1 reply

0 recast

1 reaction

Mats pfp

Just to confirm: you know that postgres can sort (correctly) on array fields such as traceAddress? If I'm not mistaken, that's what we did (when we were on postgres)

1 reply

0 recast

0 reaction

Kevin pfp

@typedarray.eth

Oh wow, I did not expect it to work out of the box on the traceAddress format. We'll have to find a way to make this work in SQLite, which (afaict) does not support array column types. New thread to pull on though - thanks!

1 reply

0 recast

1 reaction

Mats pfp

Double check that storing as a TEXT and sorting it won't work for you. I think it might produce the same ordering

1 reply

0 recast

0 reaction

Kevin pfp

@typedarray.eth

I think that will break if any elements exceed 9, eg these will not order properly: [0,1,9] [0,1,11] Thanks again, have more to play with now!

2 replies

0 recast

1 reaction

will pfp

this is a Bad Answer, but maybe it'll spark a good one: iirc EVM has a maximum call-depth of 1024, so can you just add 1024 columns and then order_by ~all of them? most of entries stay blank and most ordering should happen at the early columns so i could imagine a world where it's handled reasonably well by sqlite

2 replies

0 recast

0 reaction

Kevin pfp

@typedarray.eth

This was my first idea! And the practical max call depth is ~400 after EIP150. The real challenge is that there's no practical limit to the # of sibling calls.

1 reply

0 recast

1 reaction

will pfp

wait why does that matter? the sibling calls are differentiated by an int (or w/e you use in sqlite for bigints) which surely has enough headroom for sibling calls? what am i missing?

1 reply

0 recast

0 reaction

Kevin pfp

@typedarray.eth

Ah, yes it's fine if each depth gets a column. But, call patterns like this A CALL B ... 400 more CALLs deep CALL C CALL D ... 10000 more CALLs wide CALL Z make it difficult to do anything more compact, eg allocating 3 digits in an int to each depth.

1 reply

0 recast

0 reaction