Fix 8 bugs in proxy + exclude log files from git

[Copilot]

Multiple bugs across main.go, proxy/common.go, proxy/client.go, proxy/inputer.go, and proxy/outputer.go, plus a committed log file.

Bug Fixes

• OOB panic (main.go): Slice length equality check ran after indexing into fromaddr/toaddr/proxyproto — reordered to validate lengths first
• Dead comparison (common.go): uint32 msglen <= 0 is always false — changed to == 0 in both recvFrom and sendTo
• Data races (common.go, inputer.go, outputer.go):
  • pingflag/pongflag mixed plain pointer dereference with atomic.Add across goroutines — replaced all reads/writes with atomic.Load/atomic.Store
  • ProxyConn.pinged and ProxyConn.actived accessed unsynchronized from multiple goroutines — changed both fields to int32 and guarded all accesses with sync/atomic
• CRC check after dispatch (common.go): In recvFrom, CRC validation ran after recvch.Write(f), so corrupt frames were already consumed before detection — moved check before the write
• Silent connection leak (client.go): processLoginRsp returned without setting serverconn.needclose = true when iniService failed, leaving the connection open
• Typo (client.go): Thread group name "Clent" → "Client"

Git Hygiene

• Added *.log to .gitignore
• Removed previously committed proxy/default_DEBUG_2026-04-10.log from the index via git rm --cached

[Copilot]

Pull request overview

This PR addresses multiple runtime correctness issues in the proxy implementation (primarily concurrency safety and argument validation) and improves repo hygiene by ignoring log files.

Changes:

• Make ProxyConn.actived / pinged and ping/pong flags atomic to reduce data races; adjust message length checks and reorder CRC-check vs dispatch in recvFrom.
• Fix a client connection lifecycle issue (needclose on iniService failure) and correct a client thread-group name typo.
• Prevent CLI out-of-bounds panics by validating flag slice lengths before indexing; ignore *.log in git.

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
proxy/common.go	Switches activity/heartbeat counters and ping/pong flags to atomic operations; adjusts frame length checks and recv ordering.
proxy/inputer.go	Uses atomic increment for sonny.actived to avoid races across goroutines.
proxy/outputer.go	Uses atomic increment for sonny.actived to avoid races across goroutines.
proxy/client.go	Fixes thread group name typo and ensures failed service init triggers connection close.
main.go	Validates slice lengths before indexing into CLI flag arrays to avoid panics.
.gitignore	Adds *.log to prevent committed log artifacts.

Comments suppressed due to low confidence (1)

proxy/common.go:281

• pongtime is written in processPing and read here to populate a PONG frame, but it’s accessed via plain pointer dereference. Since sendTo runs concurrently with processPing, this is still a data race even after making pongflag atomic. Use atomic.StoreInt64 in processPing and atomic.LoadInt64 here (and keep pongtime 64-bit aligned) to make the handoff race-free.

			f = &ProxyFrame{}
			f.Type = FRAME_TYPE_PONG
			f.PongFrame = &PongFrame{}
			f.PongFrame.Time = *pongtime
		} else {

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

pingflag / pongflag clearing uses atomic.LoadInt32(...) > 0 followed by atomic.StoreInt32(..., 0). If another goroutine increments the flag between the load and store, that increment can be lost, causing missed PING/PONG sends. Consider using atomic.SwapInt32(flag, 0) (or a CAS loop) so you clear the flag and observe the previous value atomically.

[Copilot]

ProxyConn.actived / pinged were made atomic, but established and needclose are still plain bool fields that are written and read by multiple goroutines (e.g., login/open handlers vs checkPingActive/checkNeedClose). This remains a data race in Go. Consider converting them to int32 guarded by sync/atomic, or protecting the whole connection state with a mutex.