Marmalade has had some problems this week and I thought I'd explain them and my plans to fix them.
What is marmalade-repo?
marmalade is an Emacs package archive. Basically it's an artifact repository for Emacs packages. You build your package (which is often just a source file) and then upload it to marmalade. Users can download packages from marmalade by just adding it to a variable in their initialization, like this:
(add-to-list 'package-archives '("marmalade" . "http://marmalade-repo.org/packages/"))
It's easy and it's a core bit of Emacs infrastructure.
Marmalade isn't the only package repository. The FSF runs the ELPA repository, but this requires people to do copyright assignment to the FSF before contributing, so it's slow.
There is also the very excellent MELPA. This is a clever package repository that builds packages straight out of github. But MELPA works mostly for development packages right now and it only works with certain sources, and it's not an artifact repository.
How did it come to be?
A guy called Nathan built marmalade originally (he's a smart guy, check him out). But he didn't have time to look after this important thing or develop it further and so I offered to take it on.
I think it was really excellent that Nathan did the work to hand it over. It was a hassle and handing over your things is hard anyway. I know I would have found it difficult.
Nathan still owns the domain name, but I think we're all cool with that. I'd be happy to take the domain and maybe we can sort that out next year when it's up for renewal.
What are the problems?
Marmalade was built in node.js, but I think that Nathan wrote it mostly to learn how node.js works, it was very early node and a lot of it is out of date.
The immediate problem is keeping marmalade up. Something, I'm not sure what, is killing marmalade at least once a day. I was running node.js under forever which worked really well, but recently the forever depends changed to be incompatible with the node that marmalade runs under. Sigh.
So I've been running marmalade just from the command line and when it breaks the server goes down.
So that's not the smartest. I've now changed it to run inside a while loop:
while [ 1 ] ; do node marmalade ; done
So that's two problems, it's doesn't run daemonized or with any fail protection and something is crashing it.
There are a bunch of parsing and other minor problems which add up to annoyances for users or for me the maintainer, here are a few:
- there are bugs in versioning
- it's hard to change ownership of a package
- TODO the security of users is a bit weak in the datastore
- more bugs are here but even this is not exhaustive
The problem that it's kind of old node.js is considerable, it makes it hard to update and maintain. There is loads I would like to do with it but, although I like node.js, I'm no expert. And I actively dislike some of the other technologies; Nathan used his sass language for the CSS and jade for the HTML. I don't like either of those much so even changing the design of the site is a bit of a chore.
Marmalade also runs on mongo-db and I think that's a problem. Marmalade really doesn't need much database stuff, mostly it's the files/artifacts. Using Mongo is really just overhead on that. I think they should just be on the filesystem. Certainly, they'd be easier to mirror that way.
So in summary:
- (understandably) crufty implementation on aging node.js
- a bunch of specific feature bugs
- difficult to mitigate architectural choices like SASS and Mongo-Db.
What's to be done?
I am working on rewriting Marmalade in EmacsLisp using Elnode. This seems the obvious thing to do to me, Elnode works quite well (you're using it right now, to read this blog page) and can scale (using multiple Emacs instances) and can do everything we need it do... and it's EmacsLisp. It seems obvious that a package repository for EmacsLisp should be written in EmacsLisp.
The thing is, it's not a trivial rewrite and I don't have much time to do it so people are either going to have to be patient, or help fix the time.
What can other people do?
- start helping me fix Marmalade
- OR start pressing the donate button on my blog
Asking for donations feels awful to say, but if you use Marmalade you're getting something that is run for free by someone (me) who writes a lot of EmacsLisp and provides a way for you and other people to get and share that EmacsLisp in the easiest possible way. So that's the justification for asking for a few dollars here and there. Alternately, there's also a gittip option. Every little helps, if everyone who used Marmalade donated just $2 I'd easily have enough to run it on it's own server for a whole year, that would help a lot.
you can also give to Marmalade with gittip
No money! So how do I help?
If you can code EmacsLisp you could definitely help make a better Marmalade, here's the feature plan as it stands:
- make scripts to extract the packages from the existing mongo db
- this has to work iteratively so we can initially mirror the current Marmalade to make sure it works
- it's hard to give people the schema of mongo because it has important user information in it
- so I will probably have to do this
- make a new package upload handler
- in Elnode we have to:
- present an upload package form
- accept the upload
- decode the package properly
- find a directory to put the package file in and save it there
- in Elnode we have to:
- make a new package downloader
- download the file with cache headers so nginx can serve it out of cache
- output the visit as a log entry
- update an in-memory cache of the number of downloads
- this possibly has to run as a separate emacs instance (so it can scale independantly of everything else)
- also make the thing that reads the log files and reproduces the correct in-memory state
- make a new package search
- search the directory structure of packages and return hits
- make a new user system
- not storing passwords
- choosing a database (I like postgresql)
- put it all together
- a few HTML pages for the front end and the login stuff
- and we're done
The repo is on github (though it's quite out of date) but if you wanted to make issues there to claim ownership of a piece of work that would be fine.
Alternately I hang out on
#emacs on freednode a lot. You could
talk to me there about getting going.
Other stuff I'd like to do with Marmalade
I'd really like to fix the support situation. I think it would be fine if Marmalade upload died without knocking out the download. We could achieve that with mirroring relatively easily.
The reason I didn't notice that Marmalade was down was that I was away from the Internet. Being alerted would have meant that I could have fixed the problem. I really need to be alerted in future, some sort of SMS gateway would be useful.
Marmalade has lots of possibilities. Because it stores artifacts and it would be really interesting to make it easy to build into a continuous integration system, maybe with travis. I'd like to do that.
Another interesting thing would be application specific package repositories. This is a collection of just the packages that you want for a paritcular application. This would make it easy to deploy Emacs webapps, on Heroku say, with a package archive that was just the packages you needed.
The last big thing I'd like to do with Marmalade is mine the EmacsLisp there to make an index. If we could index all the Lisp there, marking what calls what, and how and maybe at several depth levels, we would have an amazing resource for looking at Lisp code. It would make it easier to learn and it would make it easier to work on and it would make for some really interesting statistics.
Update - Why not something more sensible?
Ok, if the node.js code is a bit broken and difficult to maintain, why not write this in something more sensible? PHP, Django? Rails?
Well there are a couple of reasons:
it's the Emacs community that is going to have to maintain this - if I want to give up I'm more likely to find someone else to run it and take it forward from inside the Emacs community than I am elsewhere. And the Emacs community is normally happy with Elisp, which is what Elnode is written in. It works for us, it's the most sensible thing for us.
secondly, I'm doing it - I don't have loads of time and I don't want to spend it thrashing around with a PHP app. Even a Django app. I wrote Elnode and Elnode apps are really easy for me to write. Much easier than anything else right now.
Of course, if you want to contribute a PHP app that works I'd be happy to accept it, or to hand over the running of marmalade to you if you think Elnode is a risky plan.