You can call the linter n times, but only do the work once. You know a configuration is not what you are after because it makes no sense to run that linter multiple time. This is a property of the linter, not of the environment in which it runs so we clearly are facing a new type of linter, not a new workflow to be configured

Reorder includes

rebase

Reoder includes

rebase

Address comments

Why can't the linter un once to begin with, instead o having a linter that run multiple times and then build all kind of scafolding to have it run only once ?

In D3269#77348, @jasonbcox wrote:

While mildly useful, I don't think we should be making changes like this until backports are up-to-date.

There is already logic to check if a push is minimal or not, duplicating indicate the approach is wrong.

There are still a lot of leftovers, like the flags that are passed down to functions that do not need them, the caching layer depending on flags, or the exception tot he banning code for schnorr txns. Not all need to be done in this patch, but this need some battle plan or we'll have a lot of pieces laying around.

Why not use command line tools like aspell to do this ? I think some of them are already have arc integration.

If you abstract the node in some class and call from there, you can mock it and unit test the whole thing without a node.

The elephant in the room is that this doesn't set checkpoints.