Git designed for AI agents.
// Package dontpanic provides function wrappers and supervisors to ensure
// that wrapped code does not panic and cause program crashes.
//
// When should you use this package? Anytime you are running a function or
// goroutine where it isn't obvious whether it can or can't panic. This may
// be a higher risk in long running goroutines and functions or ones that are
// difficult to test completely.
package dontpanic
import (
"sync"
"time"
"github.com/feder-cr/git-ai/internal/log"
sentry "github.com/getsentry/sentry-go"
)
// Try will wrap the provided function with a panic recovery. If a panic occurs,
// the recovered panic will be sent to Sentry and logged as an error.
// Returns `true` if no panic and `false` otherwise.
func Try(logger log.Logger, fn func()) bool { return catchAndLog(logger, fn) }
// Go will run the provided function in a goroutine and recover from any
// panics. If a panic occurs, the recovered panic will be sent to Sentry
// and logged as an error. Go is best used in fire-and-forget goroutines where
// observability is lost.
func Go(logger log.Logger, fn func()) { go Try(logger, fn) }
func catchAndLog(logger log.Logger, fn func()) bool {
var id *sentry.EventID
var recovered interface{}
normal := true
func() {
defer func() {
recovered = recover()
if recovered != nil {
normal = false
}
if err, ok := recovered.(error); ok {
id = sentry.CaptureException(err)
}
}()
fn()
}()
if id == nil || *id == "" {
return normal
}
logger.WithFields(log.Fields{
"sentry_id": id,
"recovered": recovered,
}).Error("dontpanic: recovered value sent to Sentry")
return normal
}
// Forever encapsulates logic to run a function forever.
type Forever struct {
logger log.Logger
backoff time.Duration
cancelOnce sync.Once
cancelCh chan struct{}
doneCh chan struct{}
}
// NewForever creates a new Forever struct. The given duration controls how long retry of a
// function should be delayed if the function were to thrown an error.
func NewForever(logger log.Logger, backoff time.Duration) *Forever {
return &Forever{
logger: logger,
backoff: backoff,
cancelCh: make(chan struct{}),
doneCh: make(chan struct{}),
}
}
// Go will keep retrying a function fn in a goroutine forever in the background (until the process
// exits) while recovering from panics. Each time a closure panics, the recovered value will be
// sent to Sentry and logged at level error. The provided backoff will delay retries to enable
// "breathing" room to prevent potentially worsening the situation.
func (f *Forever) Go(fn func()) {
go func() {
defer close(f.doneCh)
for {
select {
case <-f.cancelCh:
return
default:
}
if Try(f.logger, fn) {
continue
}
if f.backoff <= 0 {
continue
}
f.logger.WithField("backoff", f.backoff).Info("dontpanic: backing off before retrying")
select {
case <-f.cancelCh:
return
case <-time.After(f.backoff):
}
}
}()
}
// Cancel cancels the walking loop.
func (f *Forever) Cancel() {
f.cancelOnce.Do(func() {
close(f.cancelCh)
<-f.doneCh
})
}