Join GitHub today
GitHub is home to over 20 million developers working together to host and review code, manage projects, and build software together.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
|Failed to load latest commit information.|
Motivations: -------------------- Sexplib/typeconv (and binprot) by Jane Street are attractive, but they rely on camlp4. I don't like camlp4. I like the metaprogramming facility it offers but it has many disadvantages. For instance, because the code is generated on-the-fly, you can not grep for it. Then someone else look at your code and see a sexp_of_xxx function and has no idea where this function is. If you have already tried to understand the source code of ruby on rails, how it works, where stuff are defined, then you'll see what I am talking about. Moreover in the past camlp4 has been "fragile" to evolutions; when a new version of OCaml appeared, camlp4 sometime broke and you could not compile your code that was relying on camlp4. So I've found a in-the-middle solution where I use camlp4 to generate code (via the small script ocamltarzan.ml), and save the generated code in a file (e.g test/foo_sexp.ml), which allows then to completely remove the dependency to camlp4. Once the code has been generated, all dependencies to camlp4 can be removed. If tomorrow an incompatible new version of camlp4 arrives (e.g. camlp6 ...), your code will _still_ work, because it does not rely on the new behavior of this camlp4. It's just regular plain good ocaml code. And you can grep over code as there is no hidden code somewhere generated on the fly. All the code is there. You can even use ocamldebug to debug your generated code. And your original code does not need weird annotations like TYPE_CONV_XXX that may confuse tool that expect regular ocaml code. Example of use: --------------- Given a file 'foo.ml' containing a type 't' which you would like to have 'sexp_of_t' and 't_of_sexp' functions, as well as 'sexp_of_tlist' and 'tlist_of_sexp', just add a comment annotation after the types as in: type t = A | B (* with tarzan *) type tlist = t list (* with tarzan *) Then use my ocamltarzan (from its source directory) on this file $ ./ocamltarzan -choice sexp tests/foo.ml > tests/foo_sexp.ml The file foo_sexp.ml should now contain the 'xxx_of_sexp' and 'sexp_of_xxx' functions. To use the new services offered by those functions, you can write a use_foo.ml file such as: let x = [Foo.A;Foo.A;Foo.B;Foo.A] in let sexp = Foo_sexp.sexp_of_tlist x in let s = Sexp.to_string_hum sexp (* 'hum' mean human readable *) in print_string (s ^ "\n"); let chan = open_out "out.sexp" in output_string chan s; close_out chan; let sexp2 = Sexp.load_sexp "out.sexp" in let x2 = Foo_sexp.tlist_of_sexp sexp2 in assert (x = x2); () This should lead to this output: (A A B A) Note that once foo_sexp.ml has been generated, the only thing you really need to compile your code is the lib-sexp/ directory, which as you can see is a plain regular good ocaml library, with no camlp4 stuff involved. ocamltarzan currently supports multiple code generators, the sexp, json, value and typeof, visitor, etc. See the pa/ directory. Pro and cons of tarzan vs jane: ------------------------------- pro: - less camlp4 - less complicated to build - arguably less complicated to use, e.g. no need for the Type_conv_path stuff in the ML file - better control on the code generation as can easily customize later the generated code, for instance to not display certain things in sexp (like the cocci_tag, position, etc) - can provide a path for handling different versions, an evolutionnary format - easier to debug when there is a problem ... - can grep those generated functions; no magic. cons: - have to regenerate when change code - no Type_conv_path but have to do things manually with some module aliases - fragile if change order in .ml ? Really in sexplib they put too much things together. First the lib for just sexp, that could be splitted in its own lib (like Martin did by splitting json support in json-wheel and json-static) then the sexp_of, then the of_sexp, then the fact that their code handle also generation for signatures, then there is their extensions (sexp_option fields?), and finally that you need to use camlp4 and put many annotations in your code and to use ocamlfind in makefile, and -pp, etc. It makes the barrier of use and understanding higher. It makes it also hard to extend. If you just want a simple pretty printer for your data structures, a string_of_xxx, then they ask too much commitment. I think ocamltarzan is lighter. Note: ------- problem with mutually recursive polymorphic. Solution - duplicate functions :) have wrap_of_sexp, wrap_of_sexp_2, wrap_of_sexp_3, ... as much as needed Differences with original code: -------------------------------- Note that among other things, the file pa_sexp_conv.ml is not in the lib-sexp directory as I include there only the runtime library for sexp, not the camlp4 stuff. I put everything in a single package. I removed packaging stuff; use only one dir. removed need for findlib removed OCamlMakefile removed use of (* pp cpp *); do it manually via special handling (fun that they use cpp whereas they claim camplp4 can do everything) removed need for -pack num -> nums dependency See also the modif-orig.txt files in subdirectories. I have added a pa/pa_visitor.ml to auto generate visitors. See also pa/modif-orig.txt