go-ipfs-files 2.0 updates

[magik6k]

These are updates for changes in ipfs/go-ipfs-files#2

• go-ipfs-api, go-mfs and go-unixfs need updating to have matching files versions
• All but 1 tests pass for me
  • TestAdder_ContextCancelled fails, but probably is just racy
• Feel free to pick this up once Refactor filename - file relation ipfs/go-ipfs-files#2 is merged

[GitCop]

There were the following issues with your Pull Request

• Commit: 2ffb411
• Invalid signoff. Commit message must end with
  License: MIT
  Signed-off-by:

Code contribution guidelines for IPFS Cluster are available at https://cluster.ipfs.io/developer/contribute/#code-contribution-guidelines

---

This message was auto-generated by https://gitcop.com

[coveralls]

Coverage decreased (-0.02%) to 65.09% when pulling 11cc048 on gx/files2 into e039c71 on master.

[hsanjuan]

Hey, thanks for this PR, it will save us time.

TestAdder_ContextCancelled fails, but probably is just racy

No, I think it's due to the change. A test that never failed before, failing on both travis runs, which is relevant to the code modified by this PR... I might be wrong but I think you should look closer.

Additionally Jenkins fails:

sharding.go:141: remove shardTesting\testTree\B\big_file: The process cannot access the file because it is being used by another process.

This likely means that the readers are not being closed like before between tests (or in the same test).

Allow me to be a bit critic of the 2.0.0 change: for us, it does not simplify code, it seems to introduce some breakage and it costs us some time. Not against it, I understand it will make your life easier on some other place, but still. I really appreciate that you took the time to submit the change-set here though. Can you have another look and see why that context error is not coming back anymore? I can help you too in a few days, but I'd suggest not merging until we have figured out if the issues are in cluster's land or in files' land. It wouldn't be the first time that Cluster tests catch some issue in dependencies, it might be just cluster.

[hsanjuan]

This test should not need to adder to be interating over a SliceFile. Can you explain the change?

[magik6k]

A change in adder.FromFiles changed this behaviour slightly:

Before the change in https://github.com/ipfs/ipfs-cluster/pull/613/files?utf8=%E2%9C%93&diff=split#diff-a2c1e35d2137b0f7d1899f06c11d87b6L136 we would:

• Enter into the loop
• Check context (successfully)
• Call NextFile and process it
• Loop again
• Check context (timeout in test), return

After the change:

• Enter into the loop, calling Next
• Check context (successfully)
• Process the file
• Exit loop as Next returns false (note that we don't check context as we don't re-enter the loop)
• Exit FromFiles successfully as nothing else used the context

Strictly speaking old behaviour is more correct, but since we added the file fully in both cases, I'd argue that it makes sense to just return the result instead of timing out since we are already on the return path either way.

I needed to make this test take more than one file in order to force the loop to go around at least twice

[magik6k]

Addressed the review and made CI happier. Do you mind giving this / ipfs/go-ipfs-files#2 another pass?

[hsanjuan]

Let's leave this here until there is a final dep hash for files.

[hsanjuan]

And really, thanks a lot @magik6k !