Running Erlang w/o an OS

Robert Virding <>
Wed Feb 2 18:01:09 CET 2000


 writes:
>
>When the Java virtual machine as specified, there was some hope that
>Erlang could run in it quite easily. This was dampened a bit because
>there are some things about the JVM which make life hard (I can't
>remember what). Subsequently other people have shown that maybe it's
>not so hard after all, e.g.
>
>       Mobile Haskell: Compiling Lazy Functional Languages for the 
>	      Java Virtual Machine by David Wakeling
>
>	      http://www.dcs.ex.ac.uk/~david/research/java.htm

Unfortunately if you read the papers here you will see that while it 
possible to use the JVM the results are disappointing.  The resultant 
implementations were *slower* by a factor 10-40 than a "normal" compiled 
implementation.

While Erlang is not lazy we would have the same problems as those 
mentioned, mainly the lack of good primitives to do last call optimisation. 
 This is absolutely critical and programming to get it slows down the whole 
system.  There have been many suggestions to add hooks for LCO to the JVM 
but as far as I know none have been implemented.  This is a bit surprising 
as JAVA could make good use of them as well.

So it is not so hard if you can accept the major slow down.  We here at 
Bluetail most definitely can not.  Ulf?

	Robert





More information about the erlang-questions mailing list