PackageRank #1091
PackageRank #1091
Conversation
| versionList = sortBy (flip compare) | ||
| $ map (pkgVersion . package . packageDescription) (pkgDesc <$> pkgs) | ||
| packageEntr = do | ||
| tarB <- packageTarball tarCache . head $ pkgs |
| return (temp <> versS <> codeS <> authorScore maintainers pkg) | ||
|
|
||
| where | ||
| pkg = packageDescription <$> pkgDesc $ last pkgs |
| constructItem :: [PkgInfo] -> IO (PackageName, PackageItem) | ||
| constructItem pkgs = do | ||
| let pkgname = packageName pkg | ||
| pkg = last pkgs |
| total | ||
| . (<>) (rankPackagePage pkgD) | ||
| <$> rankIO versions recentDownloads maintainers docs env tarCache pkgs | ||
| where pkgD = packageDescription $ pkgDesc $ last pkgs |
| False -> do | ||
| index <- queryGetPackageIndex | ||
| let pkgs = PackageIndex.lookupPackageName index pkgname | ||
| case pkgs of |
There was a problem hiding this comment.
this can case on Data.NonEmpty.nonEmpty pkgs instead. Then the Just nonEmptyList case can pass that list on to constructItem
|
Thank you for the comments In my last two commits I fixed all the partial functions in PackageRank. The partial function in PackageList is in my opinion better left as is since it was that way even before my changes and on line 256 using NonEmpty would lead to Maybe PackageIndex. Which is not something that is wanted? |
|
At this moment PackageRank should be structure-wise complete. Some parameters will need to be adjusted and there may be some bugs. Caching some data (ex. freshness for dependencies would require parsing all their PkgInfo (not included)) is missing. UI is a bit arbitrary so that might also change. |
|
UI at the moment @ysangkok |
|
What's blocking us from having this merged? Looks like a great feature to have. |
|
The indirect dependencies feature was merged and it would be nice to use it here, adding a cache for the package values would be good and the parameters are somewhat arbitrary at the moment. I plan to get this into mergeable state in June. |
|
@kubaneko Good, that's great to hear. |
…ank will display as Int between 1000 and 0
|
I rebased and added some version of #1082 to PackageRank. I tested this by mirroring 600 package from hackage and the numbers it gives look reasonable. The performance implications are unclear to me, since I do not have the download speed to mirror the whole hackage but it was not too bad (I am unsure if adding indirect dependencies is a problem). There is definitely still work that can be done:
Currently I would like if someone with a hackage mirror would try what the slowdown is. If it is not noticable than I propose to merge this otherwise I will get to making the cache. Any more comments/suggestions are of course welcome |
This PR is for implementing #986 as an GSoC project. Mentored by @gbaz .
I will correct the error I made in the change.