Add support for gzip-compressed RPM files #154
Conversation
|
|
||
| -- | Extract RPM entries to a directory | ||
| extractEntries :: (PrimMonad m, MonadThrow m, MonadIO m) => Path Abs Dir -> RPMTypes.RPM -> ConduitT i o m () | ||
| extractEntries dir rpm = yield rpm .| RPM.payloadC .| decompressorFor rpm .| CPIO.readCPIO .| filterC (not . CPIO.isEntryDirectory) .| sinkDir dir |
There was a problem hiding this comment.
We're using conduit here, which allows us to do streaming decompression on the RPM entries. Some notes about reading conduit code:
Whenever you see a type like ConduitT i o m a, the type variables mean:
i- the type of elements coming from the upstreamo- the type of elements we're passing downstreamm- the underlying monad (ConduitT is a monad transformer)a- the arbitrary result type of this computation (like any other monad) -- this is not affected by the elements in the pipe
The way to read .| is to think of it as a unix pipe. Each function in between is doing a transformation on things moving down the pipe
also ignore the vomit of constraints on extractEntries; the libraries we're using can't settle on a single approach. PrimMonad is effectively the same thing as saying IO
| sendIO . runResourceT . runExceptT . runConduit $ | ||
| readRPMEntries rpmFile .| filterC (not . CPIO.isEntryDirectory) .| sinkDir | ||
| sendIO . runResourceT . runExceptT . runConduit $ do | ||
| sourceFileBS (toFilePath rpmFile) .| RPM.parseRPMC .| awaitForever (extractEntries dir) |
There was a problem hiding this comment.
Is this function in any danger of hanging with the use of awaitForever? I have seen us come across unsupported file formats when doing RPM extraction in the RPM fetcher that caused extraction to hang. I'm not aware of a better option here, but is this a valid concern? If so, it may be an acceptable failure mode for now.
There was a problem hiding this comment.
awaitForever is a bit of a misnomer: it just repeatedly tries to await input from the upstream pipe, passing it to the provided function (extractEntries dir) until no more input is available.
await in conduit doesn't actually make a thread wait or anything; everything in a conduit happens in the same thread. await is just a way to pass control to the upstream until it can give you back an element, at which point control flow moves back to you
codec-rpm'spayloadContentsCfunction only supportslzma-compressed rpm files. This adds support forgzip-compressed rpm files as well.