Caching is incompatible with @imports

[cvn]

Caching is incompatible with @imports

[jgable]

Hmm... are your imported files not run through less lint as well? Wouldn't they fail upstream or downstream?

[cvn]

Our architecture is that we have one main less files with a bunch of partials that get imported. It's not practical to lint the partials themselves because they use variables from a separate partial, and cause false positives about missing variables.

So we only lint the main file, and add the partials directory to the imports: [] config. Without caching, this works very well, including giving us accurate line numbers for linting errors in the partials. With caching on, it will ignore changes in the partials if there is no change in the main less file, and will miss any linting errors that are introduced after the cache is primed.

[jgable]

So, essentially we'd have to change the cache key of the file from just that files contents to its contents and any imports contents. I'll investigate and see what the impact is on speed.

[jgable]

This is proving to be tricky, I can build a naive solution that just parses the less into css every time and then uses the css as a cache key but that doesn't really improve performance as much as now because the less parsing is what we're trying to avoid.

Ideally, I think we'd do a regex match for import statements, grab their file contents (keep the file contents around for later lookups from other less that import it), add all the contents together as hash key.

The problems I foresee with this are relative path resolution vs passing in less options with import lookup paths.

I'm going to punt for now and probably push up something that parses every time since there is a risk of false positive cache hits right now.

[cvn]

Your idea sounds good, but I can imagine how it would be tricky. To start with, you'd have to recursively regex match all @import'ed less files for more @imports to build a complete set for hashing, right?

What about this: Read the less imports: [] option and hash the complete result set. Then compare that hash every time. It wouldn't have awareness of what files were used by the primary less files, and as a result would invalidate the cache when unrelated files changed, but it would eliminate false positive cache hits and still make things quicker for people doing most of their work in their primary less files instead of in their partials. This would only be necessary when the imports: [] array is non-empty, of course.

[jgable]

@cvn just to update on this, plan to go with your strategy of hashing the imports: [] (and possibly caching the hash based on all the names of the import globbed files). Don't really have an ETA but if I can find time today I'll do it.

[jgable]

The intermediate fix has landed in 0.19.0.

[cvn]

Very cool. Thanks, @jgable