2.6.3.4 and 2.7.0.0

[kazu-yamamoto]

2.7.0.0 should deprecate all APIs which exist in 2.6 but do not exist in 3.0. I would like to create a new 2.7-new branch based on 2.6 and then override current 2.7 in the future. 2.7-new should provide SockAddrCan since 2.7 does not provide a mechanism to extend socket addresses.

[kazu-yamamoto]

Modules to be obsoleted:

• Network
• Network.BSD

[kazu-yamamoto]

To be obsoleted in Network.Socket:

• MkSocket
• htonl, ntohl
• getPeerCred and getPeerEid
• send, sendTo, recv, recvFrom, recvLen
• inet_addr, inet_ntoa
• isConnected, isBound, isListening, isReadable, isWritable
• aNY_PORT, iNADDR_ANY, iN6ADDR_ANY, sOMAXCONN, sOL_SOCKET, sCM_RIGHTS
• packFamily, unpackFamily, packSocketType
• SockAddrCan (done in 2.6)

Need to implement:

• getPeerCredential
• ifNameToIndex and ifIndexToName

[kazu-yamamoto]

Signature changes:

API	2.7	3.0
fdSocket	Socket -> CInt	Socket -> IO CInt
mkSocket	CInt -> Family -> SocketType -> ProtocolNumber -> SocketStatus -> IO Socket	CInt -> Socket

[kazu-yamamoto]

Done in 2.6.3.4:

• Socket should be GCed.
• doc for network-uri-flag
• examples should locate in the beginning
• Do not closeFd within sendFd

[kazu-yamamoto]

@eborden Now epoch change has been merged into master. So, we can proceed this issue. Do you agree with this approach?

[kazu-yamamoto]

Note that some packages are still using Network.BSD instead of getAddrInfo. So, we should release 2.7.0.0 ASAP and wait for enough time to release 3.0.

[eborden]

@kazu-yamamoto Were there other commits in the 2.7.0.0 branch that aren't present in the 2.6.0.0? Are you just going to cherry-pick them over?

[kazu-yamamoto]

It's very hard to revert SockAddrCan stuff. So, I'm planning to cherry-pick some from the current 2.7 to 2.7-new.

But before that, we would have to release 2.6.3.4 with #288 to rescue GHC 8.2.

Cc: @qrilka

[kazu-yamamoto]

I misunderstood. 2.6.3.3 does not have #269, so #288 is meaningless.

Anyway, I believe we should release 2.6.3.4. I'm working both on 2.6.3.4 and 2.7.0.0. Please wait.

[kazu-yamamoto]

• https://github.com/kazu-yamamoto/network/tree/poc-2.6.3.4

@eborden This branch is PoC for 2.6.3.4. I think this branch is read to release. Please review

• https://github.com/kazu-yamamoto/network/tree/poc-2.7.0.0

This branch is PoC for 2.7.0.0 based on the ticket of this issue. Now I understand that we need steal getPeerCredential to deprecate getPeerCred and getPeerEid.

[kazu-yamamoto]

Should we just deprecate mkSocket and provide a function of a different name, say, toSocket :: CInt -> Socket?

[qrilka]

@eborden any estimates when new releases could be out?

[kazu-yamamoto]

@qrilka Evan is now on vacation. He will come back on 24th.

[kazu-yamamoto]

• I should back-port Unix domain #306 to 2.6.3.4 before Evan comes back.
• Optionally, I would like to resolve this: Drop include/HsNetworkConfig.h from the dist #307

[kazu-yamamoto]

aa49e91 shows that 2.6 does not have the bug of unix domain.

[kazu-yamamoto]

@eborden Would you release 2.6.3.4? No modification is necessary. Just make a tar ball and push it to hackagedb.

[kazu-yamamoto]

For 2.7, I will do:

• Rename the current 2.7 to 2.7-old
• Create a new 2.7 on the top of 2.6.
• Send a PR from my poc-2.7.0.0 branch.

[eborden]

@kazu-yamamoto Will get a candidate up today. Thanks for the reminder 😄

[eborden]

Candidate here: http://hackage.haskell.org/package/network-2.6.3.4/candidate

[eborden]

Candidate published and cafe email sent: http://hackage.haskell.org/package/network-2.6.3.4

[kazu-yamamoto]

@eborden Thank you for releasing.

[kazu-yamamoto]

Close this in favor of #315.