splithttp Read() using blocking mode

[maskedeken]

splithttp的Read()应该使用阻塞模式，不然trojan和VLESS读取第一个数据包时会返回长度为0

[mmmray]

I don't understand the reproduction steps. This hangs correctly:

func Test_readzero(t *testing.T) {
    q := NewUploadQueue(10)
    buf := make([]byte, 20)
    n, err := q.Read(buf)
    common.Must(err)
    if n == 0 {
        t.Error("n=0")
    }
}

[mmmray]

if EOF is handled here, I think line 50 can be removed? and then h.closed can be removed.

because h.pushedPackets is always closed when h.closed == true

[maskedeken]

I don't understand the reproduction steps. This hangs correctly:

func Test_readzero(t *testing.T) {
    q := NewUploadQueue(10)
    buf := make([]byte, 20)
    n, err := q.Read(buf)
    common.Must(err)
    if n == 0 {
        t.Error("n=0")
    }
}

Splithttp doesn't work with Trojan protocol, u can give it a try

[maskedeken]

I don't understand the reproduction steps. This hangs correctly:

func Test_readzero(t *testing.T) {
    q := NewUploadQueue(10)
    buf := make([]byte, 20)
    n, err := q.Read(buf)
    common.Must(err)
    if n == 0 {
        t.Error("n=0")
    }
}

it hangs here

Xray-core/transport/internet/splithttp/upload_queue.go

Line 91 in 7acd5a6

packet, more := <-h.pushedPackets

but it returns 0 directly when h.pushedPackets gets data

Xray-core/transport/internet/splithttp/upload_queue.go

Line 98 in 7acd5a6

return 0, nil

[mmmray]

thanks, i realized it after you said trojan is broken, and pushed a unittest. I could not reproduce it with VLESS, only trojan.

it seems httpupgrade might have a similar issue here (but for downloaded data)

i think in addition to this PR, trojan should also handle 0-length reads better, otherwise the issue has to be considered for N transports maybe not necessary

[mmmray]

looks good but i don't dare to merge

[yuhan6665]

Thanks guys for the great collaboration! @mmmray now that you can maintain xray-core,, you are welcomed to merge, at least with code you are familiar with :)

[yuhan6665]

Some points are:

• use squash if one change has multiple commits
• commit messages must be in English (In case of this pr, because the original title was in Chinese, it need to be re-write before squash)
• if you run into any issues, feel free to ping me