Design per-module logging system

[ALSchwalm]

Ultimately, it would be very valuable to be able to turn on verbose debugging on a per-module basis. I don't have any design in mind, just that you could set it in the mythril cfg to help debugging without being overwhelmed with other messages.

[paulcacheux]

Hi Adam, we are interested in tackling this issue. After some tests on our side it is clear to us that performance could be impacted by this filtering.

Currently what we though of was adding a configuration like:

{ 
  "version": 1, 
  "log": { 
    "*": "error", 
    "mythril::virtdev::pci": "trace" 
  }, 
  "vms": ... 
} 

Then for each log:

• We check if the module (metadata.target()) is specified (like mythril::virtdev::pci is in the example)
• We use the provided log max level if it is the case
• We use the default level (*) otherwise

This would mean that each log call now has a BTreeMap entry search and multiple string comparisons. Boot-to-shell time is increased by 50% by this change.

A solution to this problem would be to use the crate hashbrown which is an implementation of HashMap usable in a no-std environment (it was even used as the base of the rework of std’s HashMap). It improves performance in a significant way compared to BTreeMap, but is an additional dependency.

What do you think about this approach ?