From lechner@cs.uml.edu  Mon Mar  6 21:29:33 2006
From: Bob Lechner <lechner@cs.uml.edu>
Subject: EERDs to define a language? Correction: STDs capture the [LL] grammar rules.
To: sfrye@cs.uml.edu (Scott Frye)
RJLRef: $PH/06s524/SWFChptr8Notes2.txt

EERDs to define a language? Correction: 
(I forgot to add this in prior msg.)
The ERD defines a repository RDB format, a parser's
data structures, not its behavior.

Nested state machines can capture the [LL] grammar rules
for a context-free recursive descent parser.
State actions need to be able to invoke other State Machines,
even recursively. LCP can handle recursion. 


2kf522 had a parser-generator project using LCP 
which was incomplete. See

$PH/2kf522/LCPGuardConditions/2kf522LCPGC_wt_proj2k_Docs 

Here, parsing was only a sample application; the real goal was
to add guard condition pointers to LCP's State Model database,
so they could be separately compiled and executed 
like State Actions are.


I believe a parser generator could be b ased on the BNF
plus context-sensitive scoping rules, but there's no point
since Lex/Yacc or Bison is a good parser generator
(a specialized DSL in SWF terms).
(YACC/Bison's LR parser generators define state machines -
but they are too obfuscated by LR parsing semantics
to be useful as human documentation.)

OTOH, from a good C++ grammar with namescope recognition
we could parse out more than gcc lets us see, 
like variable def/ref and function calltree associations, 
for use in tools that can visualize them and help debugging.

[I expect that long before I could parse C++
(scoping rules being the hard part),
Eclipse will have such tools, if it doesn't already .
(Microsoft must have them now, in order to manage its WinXP 
and Office product families.) ]

But it takes state actions to capture the context-sensitive 
non-BNF constraints and define the code generator's translated
output. 


The EERDs define a repository that can hold multiple 
instances of a parse tree for one or many source code files.
[Like bde, they need a FIle or FileSpec or FileVersion root
entity to separate trees for distinct source code files.

[But to capture all inter-file relationships they probably need
a directory entity type above file level with multiple 
entries, like CVSROOT and make know about for BDE.]

And it takes state actions to capture the context-sensitive
non-BNF constraints and construct the objects in the code 
generator's parse tree and/or translated code output.
------------------------------------

> From sfrye@cs.uml.edu  Thu Mar  2 08:40:28 2006
> Subject: Chapter 7.[8]
>... 
> 
> What sparked my interest the most, so far, was the correlation of an 
> Abstract Syntax tree to a Metamodel.  I never made the connection that the 
> entity relationship diagrams we had been using in your class  
> could be used to define a language.
>