* Robert Engels <reng...@ix.netcom.com> [190301 15:07]:
> I am pretty sure it is the memory model unless there is other code in
> the stdlib that sets the dove back to nil - which wouldn’t make sense
> since the channel should be unique to the context instance and that
> would allow multiple done channels to be created.
I don't understand what you are saying. I absolutely agree that the
memory model is what determines whether the OP's code is correct or not.
What I was saying was that any ambiguity or lack of specificity in the
memory model is not relevant to this case. I was referring to this
statement:
* Robert Engels <reng...@ix.netcom.com> [190301 08:57]:
> This is actually a weakness due to the memory model of Go not being
> fully specified IMO.
The current memory model does not allow the compiler to move code that
is written after a synchronization point to before that synchronization
point.
Under "Happens Before" in the memory model:
Within a single goroutine, the happens-before order is the order
expressed by the program.
If you can show a reordering of the OP's code that the current memory
model is not specific enough to disallow that would cause his code to
not work as intended, then I will agree with you.
>> * dongshujin.be...@gmail.com <dongshujin.be...@gmail.com> [190301 08:50]:
>> 1 > func (c *cancelCtx) Done() <-chan struct{} {
>> 2 > c.mu.Lock()
>> 3 > if c.done == nil {
>> 4 > c.done = make(chan struct{})
>> 5 > }
>> 6 > c.mu.Unlock()
>> 7 > return c.done
>> 8 > }
I do remember a while back there was considerable discussion about
whether or not the memory model was ambiguous about certain things, and
I was left unconvinced either way; in other words, this juror is still
accepting evidence.
However, in this specific case, in a specific instance of a goroutine
whose execution has progressed to line 7, every write to c.done can
either be proved to be the assignment on line 4 of that same goroutine,
or can be proved to happen-before the lock on line 2 of that same
goroutine is acquired. The current memory model is not ambiguous about
this case.
...Marvin
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
