Path: utzoo!attcan!uunet!world!esegue!compilers-sender
From: meissner@osf.org
Newsgroups: comp.compilers
Subject: time to write a compiler
Keywords: Fortran, question
Message-ID: <9011051653.AA04571@curley.osf.org>
Date: 5 Nov 90 16:53:11 GMT
Sender: compilers-sender@iecc.cambridge.ma.us
Reply-To: meissner@osf.org
Organization: Compilers Central
Lines: 67
Approved: compilers@iecc.cambridge.ma.us
In-Reply-To: roman@gaudi.ccsf.caltech.edu's message of 31 Oct 90 18:06:32 GMT

| From: roman@gaudi.ccsf.caltech.edu (Roman Salvador)
| 
|    I would like to find out the time it takes (more or less) to write a
| compiler (i.e. a Fortran 90 one).

It depends on lots of different factors, such as:

	the skill level and experience of the person(s) doing the
	work;

	the complexity of the language proper;

	the complexity of the library (if provided);

	the tools available;

	doing 'just' a front end and using an existing back end, or
	writing the whole ball of wax;

	the target machine(s);

	the desired optimization level(s).

As another data point, I wrote the front end for the Data General MV/Eclipse
C compiler from scratch to plug into the common language code generator and
optimizer.  It took about 9 months from the start of the project until I had
a reasonable compiler (and another couple of years until it was rock solid).
I seem to recall the library took about 4-6 months for the first cut, but
since I didn't do much of this work, and we eventually rewrote a lot of it to
tie in with a UNIX emulation library developed elsewhere, I may be off on
this.  I had a free hand in writing the front end, since the previous person
on the project had left before doing much code development and I nuked what
was left.

Some things that I remember about this:

	I had to develop the compiler in PL/1, and it took some time
	to learn the language.

	I had to relearn C, since I had last used the V6 C compiler,
	and the language had changed since then.

	I was still a new programmer (I had graduated about 1-2 years
	previously).

	The parser generator that I used had some severe restrictions
	in the early days, since it was originally run in 16-bit mode,
	and could not handle 16 levels of precedence + the rest of C,
	so I had to do operator precedence in PL/1 rather than in the
	grammar.

	The reference manual (K&R) had some subtle flaws in it (you
	couldn't derive a function which returned a pointer to a
	function and took some arguments from the grammar).

	Convincing the code generator and optimizer to add new features
	that C needed (notably the ?: operator).

	Convincing the debugger to expand the common symbol table
	format, to handle user defined types, and C specific items.

--
Michael Meissner	email: meissner@osf.org		phone: 617-621-8861
Open Software Foundation, 11 Cambridge Center, Cambridge, MA, 02142
-- 
Send compilers articles to compilers@iecc.cambridge.ma.us or
{ima | spdcc | world}!esegue!compilers.  Meta-mail to compilers-request.