Post Sprint Report: LFE Experience
We are a small team of developers who are trying to get a pretty major web application launched in 12 weeks. This is a challenge in its own right, but as we were evaluating web platforms, we ended up rejecting most others in favor of Erlang. The choice was made collectively by all the devs who evaluated all of our different options (the others being PHP, node.js, Clojure, Scala, Haskell). We picked Erlang primarily for the scalability, reliability and ecosystem support.
Of the team, I am the only experienced Lisper (Scheme, Racket, CL). I was naturally attracted to LFE because of the philosophy of staying close to Erlang, but still being a Lisp. I further made the decision that at least some people would use LFE for development, perhaps mixed with Erlang. So we now have a team that is learning Erlang and LFE at the same time. The devs are all very competent, but this learning curve is our biggest risk at this point. It is my hope that LFE will help us transcend the learning curve. There are some indications that this is happening, but I will know more for sure in a few more days.
I am impressed at the correctness of the implementation. I have not seen unpredictable behavior yet and have unearthed no major correctness issues in the language implementation. Kudos to the team for that!
I like the closeness to Erlang. Being able to mindlessly call Erlang modules is a huge plus. Including .hrl files and having the records available in the defrecord form is very convenient. While I am no Erlang expert, I find converting cut and pasted Erlang code to LFE quite easy. So far, I only had a little bit of trouble with Bit comprehensions on that front. But, I suspect that is more because I am new to Erlang.
The compiler seems fast enough, but I'll know more as the number of source files grow.
The Moderately Good
Documentation: … It exists, but things are hard to find. I realize this is a work in progress and hopefully we will be able to help in some ways.
The LFE REPL: I was thankful for it for understanding behavior which was not in the documentation or if I was too lazy to look it up.
Unit Test Framework: Useful for many cases, but it took me a while to get it working correctly. I had trouble understanding test outputs, and the inability to isolate runs to specific tests was a little painful.
Please bear with me on this. Not all of this relates to LFE entirely, but I feel it better to list it here in case LFE can have better solutions. The section is bigger because I'm trying to give all the gory details.
The Macro System
About 80% of the code I wrote in the past 10 days is for macros. It was a painful experience.
The biggest problem is that the macro system does not report errors properly. If an expansion encounters an error, all that the compiler reports is something like "could not expand form". Or, even more frustratingly "bad application". No other information is provided about what the error was and where it arose. This was true in the LFE REPL as well. It would REALLY REALLY help if the underlying error is reported.
I got around these issues by writing helper functions which I could debug in the repl, but it took me a while to settle on this methodology. It makes my macros a trivial shell over a helper function, which makes for uglier code.
Note, however, the macro system works correctly once the macros are debugged. There are no issues with correctness.
My ideal situation here would be that LFE provide a mechanism for letting developers choose to always use Erlang binaries. We are currently discussing exactly this on the mail list.
Erlang's Formatted I/O
Given my long exposure to Common Lisp's format, I feel that
"~p" is woefully
inadequate. One of my resolutions is to contribute a CL compatible format
function to LFE. I would love to collaborate with anyone who is interested
One Right Way
One other thing on my wish-list would be that LFE be a little more opinionated on syntax choices. Allowing multiple ways to define functions etc., while helpful to lazy old goats like me, probably complicates the language more than necessary. It may be best left to user-defined macros to provide alternatives.
Anyhow, that is all for now. More as things develop. Congratulations on getting LFE this far.
I am working on a document called "LFE for Erlang programmers", which I will publish to this group as soon as it reaches some level of stability.