hmm… yes… I just noticed that one can use an EQL or EQ clause in the Optima patterns. No more need for the alpha-conversions, and it works just fine with these two (EQ and EQL) in the presence of an OR clause. Never mind that…
- DM > On Jul 19, 2016, at 23:58, David McClain <d...@refined-audiometrics.com> > wrote: > > Having used the NAMED-READTABLES package for several days, I can state that > it is very good for my purposes. Many thanks for pointing that one out. > > For many years I had used the #T reader macro to refer to structs in my > pattern matching compiler. Then I found CL-UNIFICATION and wanted to give it > a try. And just as surely, Marco had also invented a #T reader macro for > nearly the same purpose. > > CL-UNIFICATION is quite powerful and useful, but is much more general than > needed for pattern matching driven computations. Unification works in both > directions between patterns and data, while pattern matching needs to go only > from pattern to destructuring of incoming data. CL-UNIFICATION is also a > runtime process, instead of being compiled into efficient in-line code. > > So then I discovered a package called Optima that really is a pattern > matching compiler, much like my own, which produces efficient in-line code to > effect the pattern matching. I have to say that Optima really goes a bit > further than my own compiler. I had used rearrangement of match clauses to > gain longest common prefix matching. Optima goes further than this, taking > cues from the more advanced concepts presented in “Optimizing Pattern > Matching” by Fabrice Le Fessant and Luc Maranget. > http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.6.5507 > <http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.6.5507> > > So, I have been replacing my MATCH with that from Optima. > > But along the way, I found that Optima demands linear patterns, and cannot > handle repeated mention of variables in the pattern. At first this seemed a > serious restriction. I imagine it is a requirement of the math involved in > the Pattern Matrices mentioned in the paper by Fessant and Maranget. > > I found a way to (mostly) overcome this restriction by including an > Alpha-conversion preprocess ahead of the Optima compiler, allowing for > repeated variables in a pattern by renaming the duplicates and tacking on a > WHEN post-matching clause. This works well, except for OR patterns, which > will require something more elaborate. > > So to overcome the OR pattern problems, I’m thinking of extending Optima with > REF patterns to replace the VARIABLE nodes on the repeated symbols, giving > the repeated symbols new names by alpha-conversion and having the REF symbols > produce a comparison op. But then the question becomes - what kind of > comparison op? I have been using EQL in my alpha conversions. > > I also ran into the limitation that Optima wants proper lists as patterns and > cannot handle terminal dotted vars representing the rest of the list. This > was also easily solved by preprocessing de-sugaring ahead of its compiler. > > Since I have modified the code for Optima a bit, I moved it to a local > repository. I’m not plugged into any Git groups on the Web and so I won’t be > updating the extant repositories in any way. But maybe these enhancements to > Optima would be useful to others? > > Cheers, > > - DM >
