[erlang-questions] Debugging yecc grammars: what techniques are available?
Fri Feb 17 14:56:29 CET 2017
Thanks to all who replied on this thread, and apologies for not
getting back sooner -- more urgent things got in the way.
On Sun, Feb 05, 2017 at 11:18:39AM +0100, Mikael Pettersson wrote:
> Hugo Mills writes:
> > I'm in the process of writing a domain-specific language, using a
> > custom lexer and a yecc parser. In the process of attempting to add a
> > new feature to the grammar, I've managed to break what appears to me
> > to be a completely unrelated piece of the grammar: I'm adding a
> > structure similar to a list comprehension, and suddenly the integer
> > comparison operators have stopped working in one group of cases.
> > What approaches are there for debugging this kind of problem? The
> > best I can some up with is inspection of the .yrl source in detail,
> > but that's not helping. How else can I work out what the issue is?
> > Obviously, I could post the grammar here for inspection by more
> > experienced minds than mine, but I'd prefer to find better tooling and
> > techniques rather than mere appeal to higher authority...
> Caveat: while I have experience with various parsing techniques, including
> LALR(1), I have none with the yecc tool.
> Does yecc issue any warnings about shift/reduce or reduce/reduce conflicts?
> If it does, you need to investigate those and resolve them. With yacc and
> bison it's possible to manually inspect the LALR(1) automaton for debugging.
This turned out to be my main problem: I actually had 20 unresolved
shift/reduce conflicts, but in the yecc output, they were listed at
the top of a giant list of properly resolved conflicts (using
precedence rules), and I hadn't seen them.
I fixed up those problems, and several more that arose elsewhere in
the grammar once those had been removed (for example, duplicate
definitions of some structures like comma-separated lists). This fixed
up a bunch of my parser test failures. Then I found that the
problematic integer comparison operators had actually been only
coincidentally passing the unit tests, and that the test failures I
was seeing were correct -- so I fixed a bug in the evaluator code for
Finally, I worked out how to give function invocation a high
precedence, which has, I think, fixed the last remaining issue I had
with the grammar. (I can now write f(3)(6) and 3+g(3) and get the
correct parse trees).
I've still got some known bugs in the evaluator, but they're not
parser problems, and I know how to fix those easily enough.
Hugo Mills | "There's more than one way to do it" is not a
carfax.org.uk | commandment. It is a dire warning.
PGP: E2AB1DE4 |
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 836 bytes
Desc: Digital signature
More information about the erlang-questions