evm t8n execution log files

[winsvega]

the command I use to execute evm t8n:

/home/wins/.retesteth/t8ntool/start.sh --input.alloc /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/alloc.json --input.txs /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/txs.json --input.env /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/env.json --state.fork Istanbul --output.result /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/out.json --output.alloc /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/outAlloc.json

need to get the execution log from it running:

--trace   /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/trace.json
--trace.nomemory    /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/trace.json
--trace.nostack   /tmp/3d57a356-15f4-4ac4-9ed5-394c574debc3/trace.json

where trace.json contains tx execution log like

TXHASH: 
LOG

TXHASH: 
LOG

most cases we log only one transaction. so might need a selector argument:

--tracetxhash     (for already generated tests)
--tracetxindex    (when generating tests. tx hash is unknown)

[holiman]

This is a bit unclear to me. So here's how it works now. For each transaction, it spits out files like this: trace-0.jsonl, trace-1.jsonl etc.
I initially implemented trace-<hash>.jsonl, but realized that's not good.

• A tx may be present twice in input, but the second one does not actually execute. However, the block inclusion index is unique per transaction.
• The caller may not know the tx hash, so it' simpler to use inclusion number.

If you want trace with memory, just execute it again and specify that you want memory. Same with stack.

So as I see it, the only feature to add, would be to specify an output folder for the traces?

[holiman]

I propose that we use output.basedir, and put all output files into that directory when executing.

[winsvega]

Yes like data dir is fine.

The thing with indexes is that it is easy to mix up the index compared to tx hash.

[holiman]

Ok, how about 'trace-<index>-<hash>.jsonl' ?

[winsvega]

seems good.
So

1. output dir for the files.
2. log names. index+hash

[winsvega]

Merge?