A blog at Gallium: informal posts about research, programming practice, OCaml development, and general ramblings.

Rss feed

Namespace archeology

At the very end of 2011, and then at the very beginning of 2013, I worked for some weeks on namespaces for OCaml (what they could be, why would we need them, what would be a good solution). The resulting proposal was too complex to gather steam, so I moved on -- and never got around to making the documents publicly available. Here they are.


Fast vectorizable math functions approximations

How and why I designed my own approximated implementations of log, exp, sin and cos. The approximation error is quite small, and the functions are fast and vectorizable. For example, my logapprox function is 7.5x faster in tight loops than the icsi_log function, while being more precise.


Left-recursive versus right-recursive lists in LR parsers

Contrary to top-down (LL) parsers, which do not support left recursion, bottom-up (LR) parsers support both left recursion and right recursion. When defining a list-like construct, a user of an LR parser generator, such as Menhir, faces a choice between these two flavors. Which one should she prefer?

Two considerations guide this choice: expressiveness (which flavor leads to fewer conflicts?) and performance (which flavor leads to a more efficient parser?).

In this post, I am mainly interested in discussing expressiveness. I also comment on performance in the setting of Menhir.

As we will see, the bottom line is that neither formulation seems deeply preferable to the other.


Normalization by realizability also computes

I share with Pierre Dagand the pleasure of having finished the redaction of a small article on realizability for JFLA 15 (Journées francophones des langages applicatifs). It is a rather simple presentation of the computational content of adequacy proofs, which we hoped to understand better by exhibiting their proof terms as well-typed lambda-terms in a dependently typed meta-language.



2-or-more approximation for intuitionistic logic

I just finished the proof of a property of the simply-typed lambda-calculus that hopefully can serve as a termination argument to decide whether a type has a unique inhabitant.


The 10-pages note containing the proof is available, and all comments are warmly welcome -- in particular about an already-existing proof of this result, I don't know about any. The rest of this post is (a HeVeA rendering of) the introduction, motivating the problem and presenting the fact to be proved -- or disproved. Feel free to ask yourself whether it is true or false before looking at the full note. The answer is at the last sentence of the introduction (only in the note).

It is interesting that we don't have much intuition about whether the result is true or false -- I changed my mind several times about it, and people asked about it guessed different way. The proof, however, is relatively simple and not particularly technically involved -- but I spent a few days simplifying the concepts and the notations.


Two final versions and a Verasco release

We have just submitted the final version of two papers, presenting the state of our recent research.


Lawvere theories and monads

In this Agda file, I give an introduction to Lawvere theories and Monads. It's not exactly a "gentle" introduction because:

  1. I'm still trying to grok these things myself,
  2. I've to live up to my reputation of writing unfathomable gagaposts.

We'll start with the State monad and, briefly, a Tick monad. The game is the following: I'll present them syntactically -- through a signature -- and semantically -- through a collection of equations. I'll then show how the usual monad is related to this presentation.


Paper draft: the Essence of Ornaments

The Essence of Ornaments (pdf) is a draft of a new article, hoping to provide an accessible introduction to the theory of ornaments. Any feedback on the draft, in the comments or by email, is warmly welcome.


Ornaments in practice

I (Thomas Williams) am currently doing an internship with Pierre Dagand and Didier Rémy on ornaments, a way to relate datatypes sharing a common structure. Ornaments come from the scary world of dependent types, and the goal of my internship is to implement them in the more familiar setting of ML datatypes. We've produced a draft where we present how ornaments can be useful in practice in an ML-like programming language.


Current (and future?) state of OCamlbuild parallelization

A few days ago, Hugo Heuzard asked about the details of the problem with OCamlbuild's parallelism today on mantis: PR#5754. My mantis comment quickly grew into the blog post below.

I worked last September on better parallelization for OCamlbuild, but didn't finish the thing (I couldn't get something mergeable in time for ICFP) and got distracted with tons on other things to do since. I just uploaded the state of my branch: