(@witty-crayon-22786 I think you're best to answer this) Working on targetless formatter support: • Building on the groundwork we just laid for

lint

, We'll have the "nested" types be slightly different, but impedance-matched ◦ Namely

FmtPartitions

is now a collection of partitions of filepath+metadata, building upon lint's

Partitions

collection of partitions of <anything> ◦ Additionally

FmtRequest.SubPartition

now also holds an optional snapshot (optional because

lint

won't have the snapshot to provide) ▪︎ We can provide a

@rule_helper

for getting the snapshot:

self.snapshot if self.snapshot else await Get(...)

. Maybe we lift the underscore restriction? Or move it to

flake8

so we can

#noqa

it? However the serial nature of

fmt

is where things kinda get funky. Since

Partitions

(and therefore

SubPartitions

) can overlap in unique ways, my thought is to group

SubPartitions

by any-file-overlap. All is good so far, however when we go to serially execute the batch, the "prior" snapshot is for a potentially different set of files than the current request necessitates. So I'm imagining our snapshot populater needs to subset the "prior" snapshot on what files are valid for this subpartition (falling all the way to the very first invocation) and then merge with files we've never seen before. It's all sound, but wanted to get your (or anyone's) thoughts 🙂