Corruption of strings

[earthboundkid]

I'm seeing an issue where a Go string is sometimes corrupted when passed to darwinkit. Is there some magic thing I should do first to ensure the string isn't overwritten? I don't understand how they are supposed to interact. I tried just doing a defense strings.Clone, but somehow it didn't help and I still got corrupted memory.

[earthboundkid]

This is the code snippet:

func buildName(path string) (string, error) {
	ppath := strings.Clone(path)
	// fmt.Println("*", path, ppath) // ← if I uncomment this, I don't get corruption!!
	dateAdded, err := getDateAdded(ppath)
	fmt.Println("*", path, ppath)
	if err != nil {
		return "", err
	}
	kind := getKind(path)
	name := filepath.Base(path)
	return fmt.Sprintf("%d/%02d/%s/%s", dateAdded.Year(), dateAdded.Month(), kind, name), nil
}

func getDateAdded(path string) (time.Time, error) {
	var dateAdded foundation.Date
	defer dateAdded.Release()
	var err foundation.Error
	defer err.Release()

	s := strings.Clone(path) // Apparently, Foundation breaks strings sometimes!!
	url := foundation.URL_FileURLWithPath(s)
	defer url.Release()
	if ok := url.GetResourceValueForKeyError(
		unsafe.Pointer(&dateAdded),
		foundation.URLAddedToDirectoryDateKey,
		unsafe.Pointer(&err),
	); !ok {
		return time.Time{}, fmt.Errorf("could not read %q", path)
	}

	unixTimestamp := dateAdded.TimeIntervalSince1970()
	seconds := math.Floor(float64(unixTimestamp))
	nanoseconds := (float64(unixTimestamp) - seconds) * 1e9

	return time.Unix(int64(seconds), int64(nanoseconds)), nil
}

[earthboundkid]

I'm doing some debugging, and it seems that the reason the defensive clone doesn't work is that \x00 is getting written on top of an unrelated string before the clone happens. This seems kind of bad: Foundation is just spewing nulls into random string space that it does not own. :-/

[earthboundkid]

I tried shoving everything into an autorelease pool and I'm not seeing the same corruption, although who knows if I just pushed it somewhere else: earthboundkid/scooter@099fe29#diff-bdce60bc059a47490191bb05e74205143a001e7cd0b81c358f0a77e468ea8a6fR168

[progrium]

Sorry for delay. But yeah, if you aren't using an app framework and just using some Foundation APIs directly you'll have to manually use autorelease pools. Though, all that does is ensure any allocations with Autorelease called (anything returned by a New* constructor in Go) are released. So I'm not sure what's happening with the string, I haven't seen this before. Maybe @programmingkidx has?

[earthboundkid]

I think part of what was going on was whatever got the string expected it to be null terminated. When I changed to passing the string with a null just tacked onto the end, it still worked, which suggests something down the line was expecting a null terminated string:

var (
	ok            bool
	unixTimestamp float64
)
// It can't hurt to add a null terminator just to be safe
s := strings.Clone(path + "\x00")

// Was getting random memory corruption,
// so let's try just throwing in a pool
objc.WithAutoreleasePool(func() {
	var dateAdded foundation.Date
	var err foundation.Error
	url := foundation.NewURLFileURLWithPath(s)
	ok = url.GetResourceValueForKeyError(
		unsafe.Pointer(&dateAdded),
		foundation.URLAddedToDirectoryDateKey,
		unsafe.Pointer(&err),
	)
	if !ok {
		return
	}
	unixTimestamp = float64(dateAdded.TimeIntervalSince1970())
})

[programmingkidx]

Maybe @programmingkidx has?

Sorry I haven't seen this issue.