A brief history of Caml (as I remember it)

Caml is originally an acronym for Categorical Abstract Machine Language . It was a pun on CAM (the categorical abstract machine) and ML, the family of languages to which it belongs. The name Caml has been maintained through the evolution of the language though the present implementation has no relation with the CAM.

Caml has been designed and implemented in the Formel project at INRIA headed by Gérard Huet until 1994 and its development now continues in project Cristal.

The Formel project became interrested in the ML language in 1980-81. ML was the meta-language of the Edinburgh version of the LCF proof assistant, designed by Robin Milner. It was implemented by a kind of interpretor written in Lisp by Mike Gordon, Robin Milner and Christopher Wadsworth. LCF itself was written partly in ML and partly in Lisp. In order to be able to use the LCF proof assistant on the various systems the Formel project was using at that time (Multics, Berkeley Unix on Vax, Symbolics) , Gérard Huet decided to make the ML implementation compatible with various Lisp compilers (MacLisp, FranzLisp, LeLisp,ZetaLisp). As a member of Formel project at that time, I participated to this work which also involved Larry Paulson, working at Cambridge University. The performances of the ML implementation were improved by the addition of a compiler. Also, I added types with constructors and pattern-matching, following ideas from Robin Milner, borrowed from Hope, a language designed by Rod Burstall and David McQueen. At some point, this implementation was called Le_ML, a name which did not survive. It was used by Larry Paulson to develop Cambridge LCF and by Mike Gordon for the first version of HOL as mentioned in The Metalanguage ML (HOL documentation).

Around 1984, three facts motivated us to get even more involved in the ML developments:

This urged me to develop a new implementation of ML based on the CAM. However, the language we implemented then was not SML but ... Caml. Why? Our main reason for developing Caml was to use it for sofware developments in the Formel project and indeed, it was used for the development of the Coq system which became, after Thierry Coquand's thesis in 1985, the main aim of the project. We were reluctant to adopt a standard that could later prevent us to adapt the language to our programming needs. In particular, the Formel project developed syntax manipulation tools (Philippe LeChenadec, Michel Mauny) which appeared useful for our developments and that we incorporated into Caml. Having to synchronize with the SML team before adopting the language modifications that seemed useful for our developments would have introduced too much delay in our work and was anyway in conflict with the notion of a "standard" language which was not supposed to evolve too much. However, we tried to incorporate in Caml most of the improvements brought by SML over Edinburgh ML.

The first implementation of Caml which appeared in 1987 and was further developed until 1992 was mainly realized by Ascander Suarez who was my PhD student at the university of Paris 7. When Ascander moved to Digital Paris Research Lab in 1988, Pierre Weis and Michel Mauny, also PhD students of mine, carried on with the development and maintenance of the system. This implementation compiled Caml to LLM3 which was the virtual machine for the Le_Lisp language.

I must admit that when the Caml development started, my experience with language implementation was very limited. Relying on the LLM3 abstract machine and on the Le_Lisp memory allocation and garbage collection saved a lot of work but could not lead to high efficiency. The CAM model led to fast closure construction and good environment sharing but was poor for value accesses in environments and made optimizations difficult. It also potentially introduced memory leaks since useless values were kept in closures. Also, I had not realized that it was more important to have good performances on non functional programs than on very functional ones. Above all, I had overlooked the importance of portability and openness. In spite of these inadequacies, for which I am initially responsible, Ascander, Pierre and Michel did quite a nice work.

In the year 1990-91, Xavier Leroy designed a completely new implementation of Caml based on a bytecode interpretor written in C and Damien Doligez provided it with a very good memory allocation capability. This new implementation, known as Caml Light, was highly portable and run easily on small machines such as Macintoshes and PCs. It replaced the old Caml implementation and highly promoted the use of Caml in Education and in research teams. The stream type and the parsing facilities it offers, due to Michel Mauny, is issued from a continued effort of the Formel project to promote syntax manipulation tools. Finally, in 1996, Xavier Leroy made Objective Caml available which improves over Caml Light by its high level module facility, its object-oriented features and its native code compilers which make caml more competitive with imperative languages such as C in terms of performances. Didier Rémy should be mentioned for his influence on the objective-oriented aspects implemented by Jérô,me Vouillon and also at other moments of Caml evolution.

Besides these mainstream Caml developments, one should also mention other compilers such as BIGLOO developed by Manuel Serrano and Pierre Weis, Camlot by Régis Cridlig, CeML by Emmanuel Chailloux. An implementation for parallel machine was made by Francis Dupont and an implementation of a lazy functional language Gaml was made by Luc Maranget, both from project Para at INRIA. The Chamau implementation by Michel Mauny and Daniel de Rauglaudre offers unique syntax manipulation facilities.

I don't know if there is any lesson to learn from this fifteen years effort. Certainly, seen from 1996, the story could have been more linear. However, through essays and errors, a capacity for producing high performance portable and flexible functional language implementations has emerged in France. Caml is used in education, in research and new applications emerge such as the Web browser MMM written by François Rouaix. Many people have contributed their best to that story and credits should be given to all of them...