WIP: Cache imports #2588
WIP: Cache imports #2588
Conversation
| @@ -0,0 +1,80 @@ | |||
| {-# LANGUAGE BlockArguments #-} | |||
| module Swarm.Language.InternCache | |||
There was a problem hiding this comment.
Should add a comment with a link to the source, https://chrispenner.ca/posts/intern-cache
| t -> throwError $ ImpureImport loc (prettyText t) | ||
|
|
||
| -- | Cache file contents for import location. | ||
| fileCache :: (MonadIO m) => InternCache m (ImportLoc Import.Resolved) Text |
There was a problem hiding this comment.
Why do we need to cache file contents separately from caching processed modules?
| canonicalLoc <- resolveImportLoc (unresolveImportDir parent <//> loc) | ||
| add $ S.singleton canonicalLoc | ||
|
|
||
| sendIO (IC.lookupCached moduleCache canonicalLoc) >>= \case |
There was a problem hiding this comment.
The big problem that I can see is that this will never consider reloading a file from disk. In other words, consider the following steps:
- Start Swarm.
- Load a scenario that imports
foo/bar.sw. - Edit
foo/bar.sw. - Reload the scenario.
You would like to get the new version of foo/bar.sw, of course, but you will get the old, cached version. To get the new version you would have to completely restart Swarm.
One solution to this I have considered is to add extra data to import locations that allow us to uniquely identify them in time as well as space: a "last modified" timestamp in the case of local files, and a content hash in the case of remote files.
However, there also needs to be a mechanism to identify when we are willing to update. For example, if we are loading a scenario and encounter import https://foo seven times while recursively loading imports, we should only fetch it over the network the first time --- even if (especially if!) the remote file changes while we are still in the process of loading. However, if we then restart the scenario, we have to fetch it over the network again to see whether it changed.
2 participants
No description provided.