Issue 563:  Object terminal missing from IDL grammar (orb_revision)
Source:  (, )
Nature:  Uncategorized
Severity: 
Summary: Summary: Given the current IDL grammar, there is no way to parse any IDL containing the keyword "Object". It never appears in the grammar as a terminal symbol.

Resolution:  resolved, closed in "97
Revised Text:  Append a line to production (31) and (32) <base_type_spec>, as follows:"| <object_type>", and insert a new production between (31) and (32), as follows"(31a) <object_type> ::="Object"". also, append sentence to second para of 3.2.4 (see corresponding arc
Actions taken:
May 1, 1997: received issue
June 20, 1997: closed issue
Discussion: 

End of Annotations:=====
Return-Path: <daemon>
X-Sender: curtis@192.67.184.65
Date: Thu, 01 May 1997 12:17:32 -0400
To: issues@omg.org, orb_revision@omg.org
From: David Curtis <curtis@omg.org>
Subject: my issues
Errors-To: owner-issues@omg.org
Sender: owner-issues@omg.org
X-OMG: issues.1

Hi guys,

Somebody told me that you looked at the first one and decided it wasn't an issue.  If this is true, could you explain it to me?  I'm pretty sure it is.

Issues for CORBA Core RTF

Issue 1. Object terminal missing from IDL grammar

Given the current IDL grammar, there is no way to parse any IDL containing the keyword "Object".  It never appears in the grammar as a terminal symbol.  For example, try to parse the following IDL given the current grammar.

	typedef Object bar;

Following this chain:

(1) <specification> ::= <definition>+
(2) <definition> ::= <type_dcl> ";"
(27) <type_dcl> ::= "typedef" <type_declarator>
(28)  <type_declarator> ::= <type_spec> <declarators>

we end up trying to parse "Object" as a <type_spec>.

(29) <type_spec> ::= <simple_type_spec>
			| <constr_type_spec>
(30) <simple_type_spec>::=<base_type_spec>
			| <template_type_spec>
			| <scoped_name>
(31)<base_type_spec>::= <floating_pt_type>
			| <integer_type>
			| <char_type>
			| <boolean_type>
			| <octet_type>
			| <any_type>
(32) <template_type_spec>::=<sequence_type>
			| <string_type>
(33) <constr_type_spec>::=<struct_type>
			| <union_type>
			| <enum_type>

Object will obviously not parse as a constructed type, so from (29) we try to parse it as <simple_type_spec>.  In production (30) we can eliminate <template_type_spec> and <scoped_name>.  It is important to recognize that Object is not an identifier, it is a keyword.  If it were an identifier, it would parse as <scoped_name>, but it isn't. It won't parse as a <base_type_spec>, either.  <base_type_spec> can only produce the following terminal keywords: "float"," double", "long", "short", "unsigned long", "unsigned short", "char", "boolean", "octet", and "any".

This isn't just the case for typedefs, it happens anywhere where <type_spec> is used in the grammar, which includes struct members and union elements.  It also affects operation return and parameter types, since they are a subset of <type_spec>:

(69) <op_type_spec> ::= <param_type_spec>
			| "void"
(75) <param_type_spec>::=<base_type_spec>
			| <string_type>
			| <scoped_name>

which means that you can't parse this:

	interface boo
	{
		void bar(in Object obj);
		Object beer();
	};

This means that most of the CORBAservices IDL won't parse, in theory.

The fix is easy; add an Object terminal under <base_type_spec>:

(31)<base_type_spec>::= <floating_pt_type>
			| <integer_type>
			| <char_type>
			| <boolean_type>
			| <octet_type>
			| <any_type>
			| <object_type>
(31a)<object_type> ::= "Object"


Return-Path: <rabin@osf.org>
X-Sender: rabin@postman.opengroup.org
Date: Sun, 25 May 1997 17:45:57 -0400
To: jmischki@wheel.dcn.davis.ca.us, curtis@omg.org
From: Paul Rabin <rabin@osf.org>
Subject: issue 563: fixing IDL grammar to support "Object"
Cc: orb_revision@omg.org

Jeff, David,

David Curtis' original proposal for fixing the IDL grammar to support "Object" was to add it as a new alternative under <base_type_spec>.  During the Core RTF meeting in Stresa, Jeff Mischkinsky argued that a better solution is to add it under <scoped_name>.  After thinking about it, the original solution seems preferable.  Both solutions work for <param_type_spec> and for <type_spec>, but <scoped_name> is also used in many contexts where "Object" would be inappropriate.

	Paul

Paul Rabin                                                          tel:     +1 617 621 8873
The Open Group - Open Software Foundation           fax:    +1 617 621 0584
11 Cambridge Center                                           email: p.rabin@opengroup.org
Cambridge, MA  02142-1405
USA


Return-Path: <jmischki@wheel.dcn.davis.ca.us>
From: Jeffrey Mischkinsky <jmischki@wheel.dcn.davis.ca.us>
Subject: Re: issue 563: fixing IDL grammar to support "Object"
To: rabin@osf.org (Paul Rabin)
Date: Tue, 27 May 1997 12:36:12 -0700 (PDT)
Cc: jmischki@wheel.dcn.davis.ca.us, curtis@omg.org, orb_revision@omg.org

> 
> Jeff, David,
> 
> David Curtis' original proposal for fixing the IDL grammar to support "Object" was to add it as a new alternative under <base_type_spec>.  During the Core RTF meeting in Stresa, Jeff Mischkinsky argued that a better solution is to add it under <scoped_name>.  After thinking about it, the original solution seems preferable.  Both solutions work for <param_type_spec> and for <type_spec>, but <scoped_name> is also used in many contexts where "Object" would be inappropriate.
> 
no problem.
  i don't particularly care, how it gets fixed. The practical result will be the same--the spec will be brought into conformance with existing implementations.
    jeff
> 	Paul
> 
> Paul Rabin                                                          tel:     +1 617 621 8873
> The Open Group - Open Software Foundation           fax:    +1 617 621 0584
> 11 Cambridge Center                                           email: p.rabin@opengroup.org
> Cambridge, MA  02142-1405
> USA
> 
> 
> 


-- 
Jeff Mischkinsky                
jmischki@dcn.davis.ca.us        +1 916-758-9850
jeffm@visigenic.com             +1 415-312-5158




#563 CLOSED
Section 3.4, pages 3-9 to 3-13
Comment: Given the current IDL grammar, there is no way to parse any
IDL containing the keyword "Object". It never appears in the grammar
as a terminal symbol.
Resolution: Append a line to production (31) <base_type_spec>, as
follows: "| <object_type>", and insert a new production between (31)
and (32), as follows: "(31a) <object_type> ::= "Object"".  Also,
append a sentence to the second paragraph of 3.2.4, as follows: "The
keywork "Object" can be used as a type specifier."
Discussion: Jeff Mischkinsky proposed adding <object_type> to
<scoped_name>, instead.