Details
-
Sub-task
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
I propose splitting some of the files that have grown wild over time into files that align with more specific functionality. It's probably best to do this in a few pieces so it doesn't invalidate anyone's patches in progress. Here's what I have in mind, looking for feedback if 1) it's not worth doing for some reason 2) it will break your patch and you'd like this to wait. I'd also like to consolidate related functions, mostly protobuf helpers, that are spread around the library into dedicated files.
Targets (can split each into a separate patch):
- (done in patch 000, committed) separate the implementation of operations from async shim code in files like filesystem.cc (make a filesystem_shims.cc). The shims are just boilerplate code that only need to change if the signature of their async counterparts change.
- (done in patch 000, committed) merge base64.cc into util.cc; base64.cc only contains a single utility function.
- (done in patch 000, committed) rename hdfs_public_api.h/cc to hdfs_ioservice.h/cc. Originally all of the implementation declarations of the public API classes like FileSystemImpl were going to live in here. Currently only the hdfs::IoServiceImpl lives in there and the other Impl classes have their own dedicated files.
- split hdfs.cc into hdfs.cc and hdfs_ext.cc. Already have a separate hdfs_ext.h for C bindings for libhdfs++ specific extensions so implementations of those that live in hdfs.cc would be moved out. Just makes things a little cleaner.
- split apart various RPC code based on classes. Things like Request and RpcConnection get defined in rpc_engine.h and then implemented in a handful of files which get confusing to navigate e.g. why would one expect Request's implementation to be in rpc_connection.cc.
- Move all of the protobuf<->C++ struct conversion helpers and protobuf wire serialization/deserialization functions into a single file. Gives us less protobuf header includes and less accidental duplication of these sorts of functions.
Like any refactoring some of it comes down to personal preferences. My hope is that by breaking these into smaller patches/commits relatively fast forward progress can be made on stuff everyone agrees while things that people are concerned about can be worked out in a way that satisfies everyone.