Python Front-end to GCC

[rusi]

Mark Janssen said:
> Unattributed
> > No its not like those 'compilers' i dont really agree with a compiler 
> > generating C/C++ and saying its producing native code. I dont really believe 
> > its truely within the statement. Compilers that do that tend to put in alot 
> > of type saftey code and debugging internals at a high level to get things 
> > working in other projects i am not saying python compilers here i havent 
> > analysed enough to say this.
>  
> Hmm, well what I'd personally find interesting from a computer science
> point of view is a app that will take a language specification in BNF
> (complete with keywords and all) and output C code which is then
> compiled to an executable as normal.  This is how a front-end should
> be designed.  A middle-layer for translating common language elements
> like lists, sets, etc, could make it easy. 

Taking this starting sortie I was going to try to justify what Mark is saying.
Somewhat along the following lines.

Things like lex and yacc (and equivalents in ecosystems other than C) give a kind of holy-grail in the following sense.

When a writer of a lex/yacc spec does his thing, he does not need to think at the C level at all.  Given that executable C is produced (and ignoring mishaps/bugs like shift-reduce conflicts etc) its a very ideal situation.
The yacc-grammar (and its lex-helper) are completely declarative and yet they result in perfectly good ( O(n)) good C code.

Taking this cue from the syntax domain one can treat it as a holy grail for semantics. ie can one specify the semantics of a language completely declaratively and have a lex/yacc *analogous* magic wand and get out a compiler/interpreter etc.

Many people have pursued this holy grail but it remains elusive. Some examples:
1. Atribute grammars
2. Utrecht univs UUAG
3. Action semantics
etc

Note a key word above is "analogous"

However when I see this:


> Okay.  The purpose of BNF (at least as I envision it) is to
> produce/specify a *context-free* "grammar".  A lexer parses the tokens
> specified in the BNF into an Abstract Syntax Tree.  If one can produce
> such a tree for any given source, the language, in theory, can be
> compiled by GCC into an executable. 

all I will say is "eyes-roll"