[go-nuts] Understanding channels deadlock

On Friday, December 21, 2012 3:52:48 PM UTC+3, Alex wrote:

Hi guys,

I'm implementing some toe algorithm and ran into following strange (as for
me) problem. I'm trying to implement multithreaded algorithm but I decided
to start with one thread before because there is at least one bug in this
algorithm. But in this single-threaded approach I'm trying to use
memoization goroutine and communicate with it through channels. But program
fails with deadlock because all goroutines are in sleep mode and nobody can
read from channel, but as I can see there should be at least one goroutine
running. It fails in get_mem function. Code is here
http://play.golang.org/p/AdmlZS-yHA .

Could somebody more confident with channels point me what is wrong in this
code?

Thanks,
Alex

On Friday, December 21, 2012 3:52:48 PM UTC+3, Alex wrote:

Hi guys,

I'm implementing some toe algorithm and ran into following strange (as for
me) problem. I'm trying to implement multithreaded algorithm but I decided
to start with one thread before because there is at least one bug in this
algorithm. But in this single-threaded approach I'm trying to use
memoization goroutine and communicate with it through channels. But program
fails with deadlock because all goroutines are in sleep mode and nobody can
read from channel, but as I can see there should be at least one goroutine
running. It fails in get_mem function. Code is here
http://play.golang.org/p/AdmlZS-yHA .

Could somebody more confident with channels point me what is wrong in this
code?

Thanks,
Alex

On Sat, Dec 22, 2012 at 12:00 AM, Alex wrote:

Heh. This is strange but after a small refactoring it works. I changed
code to:
case get := <-mem_get:
res := -1
if value, ok := mem[get.index]; ok {
res = value
}
get.ch <- res

On Friday, December 21, 2012 4:42:27 PM UTC+3, Jesse McNelis wrote:

On Sat, Dec 22, 2012 at 12:00 AM, Alex <alex....@gmail.com <javascript:>>wrote:

The problem was that when there was a value in the mem map you sent two
values to the channel, but only read one.

case get := <-mem_get:
if value, ok := mem[get.index]; ok {
get.ch <- value //sent one value here
}
get.ch <- 0 //sent again here.


The fix (it's not really a fix) you got by changing
get.ch <- -1
to
get.ch <- 0

is related to this section of code.
if val := get_mem(index); val != -1 {
return val
}

That is, because you return '0' you don't call calculate again and thus
don't deadlock in get_mem() while trying to send a new channel to the
already blocked(waiting to send the second value) memoaizer().