Longstanding issues: structs & standalone Erlang

[Kostis Sagonas]

Bengt Kleberg wrote:
 > On 2006-02-15 21:01, Claes Wikstrom wrote:
 > ...deleted
 > > 1. Strip down OTP and make a base package consisting
 > >    or erts, compiler, stdlib and kernel.
 > 
 > a good idea.

So far, I have not read any extremely convincing arguments why
doing this will make Erlang more popular or easy to adopt for
serious project development. So I am not sure whether this is
such a good idea as others think it is.

One related obstacle is that currently doing this is not as
straightforward as you may think it is.

For example, "compiler + kernel + stdlib" contain calls to
functions in the following applications:

	crypto, debugger, hipe, xref

which means that if one wants to maintain the current functionality
with a guarantee of not getting "unknown function" exceptions during
runtime, either these four applications will also need to be included
or calls to them will need to be eliminated.

If the former is chosen, continuing this process until fixpoint,
requires that the following set of applications is included:

   asn1, compiler, crypto, debugger, edoc, et, gs, hipe, inets,
   kernel, observer, mnesia, mnemosyne, parsetools, runtime_tools,
   snmp, stdlib, ssl, syntax_tools, tools, webtool, xmerl

Pretty much the bulk of OTP these days, except for

	cos*, ic, megaco, mnesia_session, odbc, orber,
	os_mon, otp_mibs, pman, sasl, ssh, toolbar, and tv

The moral: software with lots of functionality often comes with
complex dependencies between its components.

Kostis