TITLE: The Role of Parsing in a Compilers Course
AUTHOR: Eugene Wallingford
DATE: April 05, 2013  3:21 PM
DESC: 
-----
BODY:
I teach compilers again this fall.  I'm looking forward to summer,
when I'll have a chance to write some code and play with some
ideas for the course.

This morning I thought a bit about a topic that pops up every time
I prep the course.  The thoughts were prompted by a tweet from
<A HREF="https://twitter.com/jcoglan">
   James Coglan</A>,
which said "Really wish this Compilers course weren't half about
parsing. ... get on with semantics."  The ellipsis is mine; James's
tweet said something about using lex/yacc to get something up and
running fast.  Then, presumably, we could get on to the fun of
semantics.

This is a challenge for my compilers course, too.  I know I don't
want to rush through scanning and parsing, yet I also wish I had
more time for static analysis, optimization, and code generation.
Even though I know the value of parsing, I wish I had equal time
for a lot of other cool topics.

<A HREF="https://twitter.com/geoffwozniak">
   Geoff Wozniak</A>'s
response expressed one of the reasons parsing still has such a
large role in my compilers course, and so many others:

<BLOCKQUOTE><EM>
Parsing is like assembly language: it seems superfluous at the
time, but provides deep understanding later.  It's worth it.
</EM></BLOCKQUOTE>

That's part of what keeps me from de-emphasizing it in my course.
Former students often report back to me that they have used their
skill at writing parsers frequently in their careers, whether for
parsing DSLs they whip up or for making sense of a mess of data
they want to process.

A current student is doing an undergrad research project that
involves finding patterns in several years of professional
basketball play-by-play data, and his ability to parse several
files of similar but just-different-enough data proved invaluable.
Of course, he was a bit surprised that corralling the data took as
much effort as it did.  Kind of like how scanning and parsing are
such a big part of a compiler project.

I see now that James has tweeted a retraction:

<BLOCKQUOTE><EM>
... ppl are RTing something I said about wishing the Compilers
course would get done with parsing ASAP.  Don't believe this any
more.
</EM></BLOCKQUOTE>

I understand the change of opinion.  After going writing a compiler
for a big language and learning the intricacies that are possible,
it's easy to reach Geoff's position: a deep understanding comes
from the experience.

That doesn't mean I don't wish my semester were twenty weeks
instead of fifteen, so that I could go deeper on some other topics,
too.  I figure there will always be some tension in the design of
the course for just that reason.
-----