dlv
dlv
针对go语言的调试器，内部架构介绍
--headless
package main
import (
"fmt"
)
func main() {
m := 120
fmt.Printf("Hello world\n")
fmt.Printf("m = %d\n", m)
}
终端1
dlv debug --headless --api-version=2 --log --listen=127.0.0.1:8181 tmp.go
client
1 直接使用官方提供的dlv作为客户端连接
dlv connect 127.0.0.1:8181
break tmp.go:8
c
2 自己使用tcp来发送json-rpc
首先，我们使用tcpdump抓一下dlv-client和dlv-rp2服务之间的报文，在tmp.go:8处下断点
-XX 能将报文转成assic可读性
-q 能精简报文
tcpdump -XX -i lo tcp and port 8181
或者 tcpdump -q -XX -i lo tcp and port 8181
提取一下有用的信息如下
localhost.44386 > localhost.8181 发送
{"method":"RPCServer.FindLocation","params":[{"Scope":{"GoroutineID":-1,"Frame":0},"Loc":"tmp.go:8"}],"id":38}
localhost.8181 > localhost.44062 回复
{"id":38,"result":{"Locations":[{"pc":4851695,"file":"/root/workspace/go/src/tmp.go","line":8,"function":{"name":"main.main","value":4851648,"type":0,"goType":0,"optimized":false}}]},"error":null}
localhost.44386 > localhost.8181 发送
{"method":"RPCServer.CreateBreakpoint","params":[{"Breakpoint":{"id":0,"name":"","addr":4851695,"file":"","line":0,"Cond":"","continue":false,"goroutine":false,"stacktrace":0,"LoadArgs":null,"LoadLocals":null,"hitCount":null,"totalHitCount":0}}],"id":39}
localhost.8181 > localhost.44062 回复
{"id":39,"result":{"Breakpoint":{"id":1,"name":"","addr":4851695,"file":"/root/workspace/go/src/tmp.go","line":8,"functionName":"main.main","Cond":"","continue":false,"goroutine":false,"stacktrace":0,"LoadArgs":null,"LoadLocals":null,"hitCount":{},"totalHitCount":0}},"error":null}
如果我再次发报文设置断点
localhost.44386 > localhost.8181 发送
{"method":"RPCServer.CreateBreakpoint","params":[{"Breakpoint":{"id":0,"name":"","addr":4851695,"file":"","line":0,"Cond":"","continue":false,"goroutine":false,"stacktrace":0,"LoadArgs":null,"LoadLocals":null,"hitCount":null,"totalHitCount":0}}],"id":39}
那么
localhost.8181 > localhost.44062 回复
{"id":39,"result":null,"error":"Breakpoint exists at /root/workspace/go/src/tmp.go:8 at 4a07ef"}
那么可以使用nc 127.0.0.1 8181直接发送以上报文
$ nc 127.0.0.1 8181
{"method":"RPCServer.FindLocation","params":[{"Scope":{"GoroutineID":-1,"Frame":0},"Loc":"tmp.go:8"}],"id":38}
{"id":38,"result":{"Locations":[{"pc":4851695,"file":"/root/workspace/go/src/tmp.go","line":8,"function":{"name":"main.main","value":4851648,"type":0,"goType":0,"optimized":false}}]},"error":null}
{"method":"RPCServer.CreateBreakpoint","params":[{"Breakpoint":{"id":0,"name":"","addr":4851695,"file":"","line":0,"Cond":"","continue":false,"goroutine":false,"stacktrace":0,"LoadArgs":null,"LoadLocals":null,"hitCount":null,"totalHitCount":0}}],"id":39}
{"id":39,"result":{"Breakpoint":{"id":1,"name":"","addr":4851695,"file":"/root/workspace/go/src/tmp.go","line":8,"functionName":"main.main","Cond":"","continue":false,"goroutine":false,"stacktrace":0,"LoadArgs":null,"LoadLocals":null,"hitCount":{},"totalHitCount":0}},"error":null}
{"method":"RPCServer.FindLocation","params":[{"Scope":{"GoroutineID":-1,"Frame":0},"Loc":"tmp.go:8"}],"id":38}
{"id":38,"result":{"Locations":[{"pc":4851695,"file":"/root/workspace/go/src/tmp.go","line":8,"function":{"name":"main.main","value":4851648,"type":0,"goType":0,"optimized":false}}]},"error":null}{"method":"RPCServer.CreateBreakpoint","params":[{"Breakpoint":{"id":0,"name":"","addr":4851695,"file":"","line":0,"Cond":"","continue":false,"goroutine":false,"stacktrace":0,"LoadArgs":null,"LoadLocals":null,"hitCount":null,"totalHitCount":0}}],"id":39}
{"id":39,"result":null,"error":"Breakpoint exists at /root/workspace/go/src/tmp.go:8 at 4a07ef"}
自己调试自己
dlv debug /root/workspace/go/src/github.com/derekparker/delve/cmd/dlv/main.go -- debug --headless --api-version=2 --log --listen=127.0.0.1:8181 tmp.go
多文件调试
dlv debug 并不支持main package中含有多个文件的情况，之前观赏gor的代码发现的这种情况
所以交了一次pull request
通过修改过后的dlv，调试多文件 gor 并附带参数
find . -maxdepth 1 -name "*.go" | awk -F '/' '{s=s" "$2}END{print "./testdlv debug --headless --api-version=2 --log --listen=127.0.0.1:8181 " s " -- --input-raw :8000 --output-stdout"}' |bash
架构
dlv的命令主要有 dlv debug dlv test dlv trace
所有命令行的入口都是在cmd/dlv/commands.go(采用的package是spf13的cobra，相信玩vim都知道这个人)
就拿debug来看，首先会根据代码生成一个名为debug的临时文件（调试完成后defer remove掉）
传到service/rpccommon/server.go里面创建一个server变量（rpccommon.NewServer）
server.run运行debug文件，注意这里面涉及到ptrace（linux下的情况）去跟踪程序（这里可以用ps命令看一下进程处于trace状态）
dlv的上层都是做一些逻辑（service目录）行为，最底层最核心的都是在其仓库pkg的目录使用系统级别提供的进程调试工具，例如linux的ptrace
而dlv connetct也只是其client功能，所有client和srever端服务都是通过JSON-PRC完成的