sounds like a great plan 👍

nice approach for this! basically, this would handle the cases below;

docker run --rm myname:/foo ...

# note: no name specified during creation of the volume
# however, the generated name is passed on to docker run
# so it's considered a "named" volume, and won't be removed
VOL=$(docker volume create); docker run --rm -v $VOL:/foo ..

# same here, but no intermediate variable
docker run --rm -v $(docker volume create):/foo ...

In all those cases, the volume is not removed when the container is removed. However, the examples below treat the volume as ephemeral and remove the volume together with the container;

docker run --rm -v /foo ..

# image that has a `VOLUME` specified;
docker run --rm something

We may need to update some documentation explaining this

@thaJeztah I'll take a look at the docs tomorrow.

@cpuguy83 👍 just mentioning it to prevent trigger-happy merges 😄

@cpuguy83 I tested your branch. This scenario does not work, the volume still be remove if the named volume is from --volumes-from CONTAINER and the original CONTAINER has been removed.

$ docker volume create --name data
data
$ docker run -ti --name foo -v data:/data busybox
/ # exit
$ docker run -ti --rm --name bar --volumes-from foo busybox
/ #

Remove container foo
$ docker rm foo foo
And then kill container bar, the volume data will be removed

DEBU[0637] DELETE /v1.23/containers/7b4e2b232d6269036b1bf28f9b563426b0b7cac60e0125446d420a6eaca45e6e?v=1
DEBU[0637] Removing volume reference: driver local, name data

seems we should also handle https://github.com/docker/docker/blob/master/daemon/volumes.go#L73

@coolljt0725 Nice catch! Updated with a fix and another test.

code LGTM. Still considering docs additions, I assume?

We're considering this a bug fix, added to the milestone

@estesp yes, @cpuguy83 was going to work on docs today

LGTM

Rebased and added docs

Thanks @cpuguy83 can you also update the note in this section; https://github.com/docker/docker/blob/master/docs/reference/run.md#clean-up---rm ?

@thaJeztah Updated.

Thanks! Docs LGTM

ping @vdemeester @MHBauer for docs review

Updated -- also fixed windows test (hopefully, we'll see).
@vdemeester PTAL at the doc change

Docs LGTM 🐹
let's wait for green :)

docs nit, but otherwise LTGM. reading the tests helped me understand. The two new ones are simple and good for understanding the behavior. The full cases they represent might make better examples than the given examples that are present, but that is a balance between verbosity and education. It would be best if we could somehow link the code directly into the docs, but I don't think that's feasible.

@MHBauer thanks, fixed the mistake.

Hit this one on gccgo #19368

And weird message on win2lin;

22:22:29 INFO: Commmit hash is d17261b
22:22:29 INFO: Attempting curl to ping the DIND daemon
22:22:29 ERROR: Invalid syntax. Default option is not allowed more than '1' time(s).
22:22:29 Type "TIMEOUT /?" for usage.

gccgo hit again #19368... merging

Added changelog; it's a notable bug fix