mgmt/docs/release-notes/0.0.24

I've just released version 0.0.24 of mgmt!


> 600 files changed, 13622 insertions(+), 6907 deletions(-)

This is a huge and hugely important release! It has been a long time
coming. We have lambdas! I could not have done this without the
unrelentingly supportive and brilliant Samuel Gélineau.

Getting past the blockers and tricky code changes was all thanks to
him. Cleanups, polish and making it more golang idiomatic was my doing.
(The easy stuff.) If Sam wanted to be a golang expert, he could have
done it all, but it was more sensible than I do all the mundane and
filler stuff.

With that, here are a few highlights from the release:

* We have working lambdas, including iter.map =D

* We have a new function engine!

* We have improved type unification!

* We have an improved resource engine!

* We have improved many of the internal API's.

* We have so many tests.

And much more...

DOWNLOAD

Prebuilt binaries are NOT available here for this release:
https://github.com/purpleidea/mgmt/releases/tag/0.0.24

NEWS

* Our main tests are now in the excellent .txtar format. This makes
things much easier to manage.

* The iter.map function can be named as such due to parser tricks! No
need to name it xmap anymore!

* The unification solver has been greatly improved. We can infer a lot
more about function types.

* The resources API uses the context package for closing Watch and
returning early from CheckApply. The next step would be to remove all
the resource-specific timeout code and make that a metaparam.

* A new sync primitive has been added in case you'd like to use it
somewhere. We'd love help adding an even more complex one. Look in
util/sync.go for more information.

* Sam added some beautiful type inference debugging that makes things
easier for those familiar with the standard literature.

* An important bug in standalone etcd has been fixed. While embedded
etcd and automatic clustering isn't "supported" (it's buggy) the
status-quo of using your own etcd cluster is stable, and you can even
use the embedded etcd server in standalone mode...

* This means you can run `mgmt etcd` and get the standard etcd binary
behaviour that you'd get from running `etcd` normally. This makes it
easy to use both together since you only need to transport one binary
around. (And maybe mgmt will do that for you!)

* I fixed and cleaned up some sketchy code in the resource engine. I
had been unmotivated to fix this for a while because I really wanted
lambdas first, but now that they are in, I took a good look at the
code, made some fixes, and I'm really happy with it now.

* Metaparams are appropriately stateful between graph switches now.
Retry is the easy example. Limit/Burst need to be ported if you care
about these fine details.

* A RetryReset metaparam has been added. This is another good example
of how powerful metaparams are and how much potential there is for
building systems with these and future ones.

* A bunch of internal API's have been updated. This makes it better for
function and resource writers! Some GAPI changes also got pushed
through that make things clearer for those reading code.

* We have a listlookup function. It's still missing syntactic sugar
though!

* Our new function graph engine is called "dage". I think it's pretty
clever. There's a chance there is still a bug inside, but it's unclear.
Please report any issues. If you have some large machines I can test
very large and fast graphs on, please let me know.

* Lambdas really work now! The txn and ref/gc code is pretty fantastic.

* Many bugs have been killed!

* We're looking for help writing Amazon, Google, DigitalOcean, Hetzner,
etc, resources if anyone is interested, reach out to us. Particularly
if there is support from those organizations as well.

* Many other bug fixes, changes, etc...

* See the git log for more NEWS, and for anything notable I left out!

BUGS/TODO

* Function values getting _passed_ to resources doesn't work yet, but
it's not a blocker, but it would definitely be useful. We're looking
into it.

* Function graphs are unnecessarily dynamic. We might make them more
static so that we don't need as many transactions. This is really a
compiler optimization and not a bug, but it's something important we'd
like to have.

* Running two Txn's during the same pause would be really helpful. I'm
not sure how much of a performance improvement we'd get from this, but
it would sure be interesting to build. If you want to build a fancy
synchronization primitive, then let us know! Again this is not a bug.

TALKS

I don't have anything planned until CfgMgmtCamp 2024. If you'd like to
book me for a private event, or sponsor my travel for your conference,
please let me know.

MISC

We're still looking for new contributors, and there are easy, medium
and hard issues available! You're also welcome to suggest your own!
Please join us in #mgmtconfig on Libera IRC or Matrix, or ping this
list if you'd like help getting started! For details please see:

https://github.com/purpleidea/mgmt/blob/master/docs/faq.md#how-do-i-con
tribute-to-the-project-if-i-dont-know-golang

Many tagged #mgmtlove issues exist:
https://github.com/purpleidea/mgmt/issues?q=is%3Aissue+is%3Aopen+label%
3Amgmtlove

Although asking in IRC/matrix is the best way to find something to work
on.

MENTORING

We offer mentoring for new golang/mgmt hackers who want to get
involved. This is fun and friendly! You get to improve your skills,
and we get some patches in return. Ping me off-list for details.

THANKS

Thanks (alphabetically) to everyone who contributed to the latest
release:
James Shubin, Kaushal, Laurent Indermuehle, Ofek Atar, Samuel Gélineau
We had 5 unique committers since 0.0.23, and have had 88 overall.
run 'git log 0.0.23..0.0.24' to see what has changed since 0.0.23

Happy hacking,
James
@purpleidea