feat: match before checking cache

[brianmcgee]

This changes the flow of processing to make unmatched behaviour more consistent.

Before, we had been:

• traversing the filesystem
• comparing with the cache and only emitting files which had changed
• applying the matching rules to determine which formatters should be applied to a given file
• applying the formatters

Now, we do the following:

• traverse the filesystem
• apply the matching rules to determine which formatters should be applied to a given file
• compare with the cache and only emit files which have changed for formatting
• apply the formatters

It does mean we are applying the matching rules against files which we may not have to format, but in testing against Nixpkgs the performance impact appears negligible.

This makes sense since most of the processing time will be spent in the formatters, not applying some globs to file paths.

Close #433

[brianmcgee]

I just realized I can further simplify the stats since matched and formatted will always be the same.

[jfly]

This looks great, thanks for making the change =)

I just realized I can further simplify the stats since matched and formatted will always be the same.

Sorry, I think I'm not understanding the vocabulary. Here's what I thought:

• matched: the number of files in the repo that treefmt "has opinions" about
• formatted: of the files we matched, the number of files that treefmt invoked a formatter on. caching means we may choose not to format a matched file (because we know it hasn't changed).
• changed: of the files we formatted, how many of them actually materially changed

[brianmcgee]

• matched: the number of files in the repo that treefmt "has opinions" about
• formatted: of the files we matched, the number of files that treefmt invoked a formatter on. caching means we may choose not to format a matched file (because we know it hasn't changed).
• changed: of the files we formatted, how many of them actually materially changed

This is a good summary.

[jfly]

Sorry, I didn't get to my point, haha:

since matched and formatted will always be the same.

Is this true? When we have cache hits, won't we have files that we match that we end up not formatting?

[brianmcgee]

Sorry, I didn't get to my point, haha:

since matched and formatted will always be the same.

Is this true? When we have cache hits, won't we have files that we match that we end up not formatting?

You're not wrong. Now that I look at this again, I have no idea why I removed the distinction. I think I need to take a break from treefmt for a few days after this.

Putting it back in.

[jfly]

LGTM!