[clisp-list] modules initialization, package and crashes

[Hoehle, Joerg-C. <Joe...@t-...>]

Hi,

CLISP's external modules are initialized very early while starting
up. For instance,
o init_streamvars() has not yet been called (setting *TERMINAL-IO* =
etc.)
o init_packages() has not yet been called when starting without -M
o and many others as well...

CLISP catches the attempt by modules to define Lisp functions
(e.g. LISPFUN) in an unknown package (spvw.d:init_other_modules):
 { var object pack =3D =
find_package(asciz_to_string(packname,O(internal_encoding)));
   if (nullp(pack)) # Package nicht gefunden?
   { asciz_out_ss(GETTEXTL("module `%s' requires package %s." =
NLstring),
                  module->name, packname);
-> "module `mtest' requires package TESTCI."

But it cannot do so for arbitrary objects, e.g. the symbol
TESTCI::DECIDER that Peter Wood tries to use. It attempts to read them
via a string-input-stream, so a package error is signaled. As
init_streamvars() has not yet been called, a suberror follows about
*TERMINAL-IO*. Why it crashes later on I don't know - it should not.

It's not just a package problem. Any error signaled while reading the
object_tab to initialize an object causes CLISP to die.

I don't know why in the LISPFUN case an error is signaled instead of
just creating a virgin package. Maybe because CLISP was not
initialized far enough, maybe because that would have disturbed later
FIND-PACKAGE checks in Lisp code.

I don't know why init_other_modules_2() (esp. stream_read) etc. are
called that early in the process.

There may be excellent reasons for all these. Only Bruno Haible could =
tell.

I don't know what to do yet.

Regards,
	J=F6rg H=F6hle.