hannahhoward
commented
3 years ago Hi @adlrocha -- I am going to start working on this shortly
Wanted to run some stuff by you before I do: As I see it, the current test cases are broken up covers a few different concerns:
- Type of host constructed (currently raw tcp socket, Bitswap + Libp2p, full IPFS host)
- Node topology (full connection between nodes, partial connection with passive nodes, etc)
- Data fetching method (normal vs waves, UnixFX vs Graphsync)
My thinking is this:
- the only thing that should constitute a truly different type of test case is the type of host
- we out to be able to parameterize the sparse and waves test with the general IPFS transfer test, based on supporting paramaterized node topology and fetching methods
- I'd actually like to match fetching method only cover who asks who for what and when -- and move the graphsync test to be a raw libp2p+graphsync test. That means:
- TCP = raw transfer,
- Raw Bitswap = Walk method in utils (though I think we should use https://github.com/ipfs/go-merkledag/blob/master/merkledag.go#L377 which will enable us to add concurrency),
- Graphsync = graphsync.Request
- IPFS = UnixFS().Get
All of these tests share a ton of common steps, that I'd like to refactor to be as deduplicated as possible:
- setting up a data store
- putting data in the datastore for seeders
- building a libp2p host (all except TCP)
- connecting nodes in some kind of topology
- fetching in some fashion from the other nodes
- waiting for things to finish
I'd love it if the test cases themselves were quite short -- right now they're hundreds of lines and hard to read to understand what's going on, plus they copy each other a lot.
adlrocha
For the sake of development speed and rapid result explorations, there is a lot of redundant code in test cases. This needs to be fixed before it becomes "technical debt".