OCaml Weekly News

Previous Week Up Next Week


Here is the latest OCaml Weekly News, for the week of November 03 to 10, 2020.

Table of Contents

OCaml native library to draw, scale, rotate shapes on screen

Yoann asked announced

Is there any library to easily draw shapes (circle, rectangles, etc.) and rotate or scale them, and that also supports some alpha layer.

I know Graphics, but it does not look like you can rotate/scale/alpha easily shapes. I know Cairo but it depends on lablgtk to draw things on the screen and I don't need any widgets and I'd rather avoid dependencies to gtk.

For context I'm trying to port elm-playgound which allows to super easily code pictures, animations, or games (see https://package.elm-lang.org/packages/evancz/elm-playground/latest/ for its API). I am almost done porting it in a Web context thanks to js_of_ocaml, ocaml-vdom, and the SVG support in a browser, but I'd like now to port it in a native context.

Armael replied

I'm using the ocaml-cairo in a project, and it works relatively well. The library is also very nicely documented (there's an ocaml translation of the cairo tutorial, and various examples). Note that it corresponds to the cairo2 package in opam, which does not depend on lablgtk! (the downside being, I think, that the API provided by the cairo ( C) library is very imperative and can be a bit tricky to use at times).

I also know of the vg library for vector graphics, which has the virtue of being pure ocaml code. It is also very well documented. I suspect that it has a smaller feature set and number of supported backends than Cairo, but has a nicer API. Actually, it seems that you can use Cairo as a backend for Vg (https://erratique.ch/software/vg/doc/Vgr_cairo/index.html), which might give you the best of both worlds, in the case where you want to use Vg with a backend only supported by Cairo.

Igarnier added

You an use e.g. Cairo.Image.create_for_data32 to create a Cairo RGBA surface from a bigarray (that can be then shared with Sdl or whatever other lib for displaying).

Yoann Padioleau then replied

Oh you're right. I just found this example to transfer pictures from cairo to graphics: https://github.com/Chris00/ocaml-cairo/blob/master/examples/graphics_demo.ml

Armael added

Oh, and I should note that I know of a third option: https://github.com/let-def/wall

It also has a nice vector-graphics API, and can render to Tsdl (see the example). (downside: it's maybe a bit less mature and less documented than the other two, but seems to correspond to what you're looking for.)

Florent Monnier said

You can scale and rotate with SDL2 and it's bindings. Use the function Render.copyEx to copy a sprite on screen. Use the angle parameter to rotate and use rectangles of different width and height to scale. SDL2 doesn't permit to draw much shapes though, only rectangles by default, because it's made for gamedev, but you can use the SDL2_gfx lib in addition with SDL2, with its template bindings OCamlSDL2_Gfx.

Florent Monnier later added

SFML and its bindings ocaml-sfml allow more shapes drawing than SDL2.

You will probably be able to do everything you want with it, any shapes, rotate, scale, translate and more! Have a look at its api-doc, and in particular the SFShape module.

(Also please notice that SDL and SDL2 are 2 different libs, that don’t share any code, and have 2 different licenses, and with different implementations, SDL 1.2 uses software render, and SDL2 uses OpenGL inside.)

GuaCaml.0.02 : A Generic Unspecific Algorithmic in OCaml

Joan Thibault announced

GuaCaml is a simple extension to the Standard OCaml Library (Stdlib).


Snowflake : a Generic Symbolic Dynamic Programming framework

Joan Thibault announced

I have the pleasure to announce the to announce the first opam release of Snowflake : a Generic Symbolic Dynamic Programming framework interfacing WAP, RBTF and MLBDD.

Complex systems (either physical or logical) are usually structured and sparse, that is, they are build from individual components linked together, and any component is only linked to a rather small number of other components with respects to the size of the global system.

RBTF exploits this structure, by over-approximating the relations between components as a tree (called decomposition tree in the graph literature) each node of this tree being a set of components of the initial systems. Then, starting from leaves, each sub-system is solved and the solutions are projected as a new constraints on their parent node, this process is iterated until all sub-systems are solved. This step allows to condensate all constraints into a single sub-system and check their satisfiability. We call this step the Forward Reduction Process (FRP).

Finally, we can propagate all the constraints back into their initial sub-system by performing those same projection in the reverse direction. That is, each sub-system update its set of solution given the information from its parent then send the information to its children sub-systems (possibly none, if its a leaf). We call this step the Backward Propagation Process (BPP).


Using CLI arguments through modules and configuration files

erhan gundogan asked

I was looking for common approaches and best practises regarding to CLI argument passing and configuration files. I couldn't find much after googling so here is my questions:

  1. I am getting arguments with cmdliner library and passing those arguments through functions. And sometimes I have to pass these arguments through multiple functions. I am not sure if that's the correct way of doing it. Let's say I have verbose flag defined and I have logs in various modules. So how can I receive this flag whenever needed? Could you please share any resources regarding to saving arguments and accessing them from different modules?
  2. Is there a configuration file concept in OCaml? Files such as .yaml and .json

Craig Ferguson replied

For the specific case of logging, my advice is to use the Logs library or – failing that – to just copy the approach that it takes verbatim. Logs looks after some global mutable state that contains the current logging level of the program, so you don't have to bother with propagating this information throughout your program. When I'm reading OCaml code, I'm not concerned with whether any particular function might emit log lines, so I don't need this to be made painfully obvious at each call-site. If you do use Logs, it comes pre-packaged with Cmdliner specifications for setting the logging level in the logs.cli package (example here).

Generally, there are several options for propagating state throughout an OCaml program. In roughly decreasing order of explicitness:

  • pass all params explicitly to the functions that need them, precisely as you're doing right now.
  • pack params into a "context" record (or object) that is passed explicitly where it's needed. (c.f. Dune's Context and Super_context.)
  • pack params into a "context" module that is then used to instantiate functors elsewhere in your program. (c.f. Ppxlib.Ast_builder as a way of propagating a ~loc flag everywhere.)
  • use global mutable state, as in Logs.

I've seen all four of these used sensibly in OCaml programs; the best one will depend on your particular application requirements / how much you care about tracking which part of the program use which arguments.

Regarding your second point, AFAIK there's no generic library for managing config files in OCaml (i.e. what Cosmiconfig provides for NPM). Every OCaml library that I've seen that uses one tends to roll their own logic for it. You could of course use Yojson or OCaml-Yaml to read a file in one of those formats, but you'll end up managing the details yourself. The lightweight approach is to use environment variables, since Cmdliner will handle that boilerplate for you; managing config files is a pain, particularly w.r.t. things like respecting XDG_CONFIG and it's analogues on Windows.

Xavier Leroy also replied

Josh Berdine said

If you want to use globals with cmdliner, you might find it convenient to use some code such as:

(** Extension of Cmdliner supporting lighter-weight option definition *)
module Cmdliner : sig
  include module type of Cmdliner

  val mk : default:'a -> 'a Term.t -> 'a ref
  (** [mk ~default term] is a ref which, after [parse] is called, contains
      the value of the command line option specified by [term]. *)

  val parse : Term.info -> (unit -> unit Term.ret) -> unit
  (** [parse info validate] parses the command line according to the options
      declared by calls to [mk], using manual and version [info], and
      calling [validate] to check usage constraints not expressible in the
      [Term] language. *)
end = struct
  include Cmdliner

  (** existential package of a Term and a setter for a ref to receive the
      parsed value *)
  type arg = Arg : 'a Term.t * ('a -> unit) -> arg

  (** convert a list of arg packages to a term for the tuple of all the arg
      terms, and apply it to a function that sets all the receiver refs *)
  let tuple args =
    let pair (Arg (trm_x, set_x)) (Arg (trm_y, set_y)) =
      let trm_xy = Term.(const (fun a b -> (a, b)) $ trm_x $ trm_y) in
      let set_xy (a, b) = set_x a ; set_y b in
      Arg (trm_xy, set_xy)
    let init = Arg (Term.const (), fun () -> ()) in
    let (Arg (trm, set)) = List.fold_right ~f:pair args ~init in
    Term.app (Term.const set) trm

  let args : arg list ref = ref []

  let mk ~default arg =
    let var = ref default in
    let set x = var := x in
    args := Arg (arg, set) :: !args ;

  let parse info validate =
    match Term.eval (Term.(ret (const validate $ tuple !args)), info) with
    | `Ok () -> ()
    | `Error _ -> Caml.exit 1
    | `Help | `Version -> Caml.exit 0

Multicore OCaml: October 2020

Anil Madhavapeddy announced

Welcome to the October 2020 multicore OCaml report, compiled by @shakthimaan, @kayceesrk and of course myself. The previous monthly updates are also available for your perusal.

OCaml 4.12.0-dev: The upstream OCaml tree has been branched for the 4.12 release, and the OCaml readiness team is busy stabilising it with the ecosystem. The 4.12.0 development stream has significant progress towards multicore support, especially with the runtime handling of naked pointers. The release will ship with a dynamic checker for naked pointers that you can use to verify that your own codebase is clean of them, as this will be a prerequisite for OCaml 5.0 and multicore compatibility. This is activated via the --enable-naked-pointers-checker configure option.

Convergence with upstream and multicore trees: The multicore OCaml trees have seen significant robustness improvements as we've converged our trees with upstream OCaml (possible now that the upstream architectural changes are synched with the requirements of multicore). In particular, the handling of global C roots is much better in multicore now as it uses the upstream OCaml scheme, and the GC colour scheme also exactly matches upstream OCaml's. This means that community libraries from opam work increasingly well when built with multicore OCaml (using the no-effects-syntax branch).

Features: Multicore OCaml is also using domain local allocation buffers now to simplify its internals. We are also now working on benchmarking the IO subsystem, and support for CPU parallelism for the Lwt concurrency library has been added, as well as refreshing the new Asynchronous Effect-based IO (aeio) with Multicore OCaml, Lwt, and httpaf in an http-effects library.

Benchmarking: The Sandmark benchmarking test suite has additional configuration options, and there are new proposals in that project to leverage as much of the OCaml tools and ecosystem as much as possible.

As with previous updates, the Multicore OCaml ongoing, and completed tasks are listed first, which are followed by improvements to the Sandmark benchmarking test suite. Finally, the upstream OCaml related work is mentioned for your reference.

Multicore OCaml

  • Ongoing
    • ocaml-multicore/ocaml-multicore#422 Simplify minor heaps configuration logic and masking

      The PR is a step towards using Domain local allocation buffers. A Minor_heap_max size is used to reserve the minor heaps area, and Is_young for relying on a boundary check. The Minor_heap_max can be overridden using OCAMLRUNPARAM environment variable.

    • ocaml-multicore/ocaml-multicore#426 Replace global roots implementation

      An effort to replace the existing global roots implementation to be in line with OCaml's globroots. The objective is to also have a per-domain skip list, and a global orphans when a domain is terminated.

    • ocaml-multicore/ocaml-multiore#427 Garbage Collector colours change backport

      The Garbage Collector colour scheme changes in the major collector have now been backported to Multicore OCaml. The mark_entry does not include end, mark_stack_push resembles closer to trunk, and caml_shrink_mark_stack has been adapted from trunk.

    • ocaml-multicore/ocaml-multicore#429 Fix a STW interrupt race

      The STW interrupt race in caml_try_run_on_all_domains_with_spin_work is fixed in this PR, where the enter_spin_callback and enter_spin_data fields of stw_request are initialized after we interrupt other domains.

  • Completed
    • Systhreads support
      • ocaml-multicore/ocaml-multicore#381 Reimplementing Systhreads with pthreads (Domain execution contexts)

        The re-implementation of Systhreads with pthreads has been completed for Multicore OCaml. The Domain Execution Context (DEC) is introduced which allows multiple threads to run atop a domain.

      • ocaml-multicore/ocaml-multicore#410 systhreads: caml_c_thread_register and caml_c_thread_unregister

        The caml_c_thread_register and caml_c_thread_unregister functions have been reimported to systhreads. In Multicore OCaml, threads created by C code will be registered to domain 0 threads chaining.

    • Domain Local Storage
      • ocaml-multicore/ocaml-multicore#404 Domain.DLS.new_key takes an initialiser

        The Domain.DLS.new_key now accepts an initialiser argument to assign an associated value to a key, if not initialised already. Also, Domain.DLS.get no longer returns an option value.

      • ocaml-multicore/ocaml-multicore#405 Rework Domain.DLS.get search function such that it no longer allocates

        The Domain.DLS.get has been updated to remove any memory allocation, if the key already exists in the domain local storage. The PR also changes the search function to accept all inputs as variables, instead of a closure from the environment.

    • Lwt
      • ocaml-multicore/multicore-opam#33 Add lwt.5.3.0+multicore

        The Lwt.5.3.0 concurrency library has been added to support CPU parallelism with Multicore OCaml. A blog post introducing its installation and usage has been written by Sudha Parimala.

      • The Asynchronous Effect-based IO builds with a recent Lwt, and the HTTP effects demo has been updated to work with Multicore OCaml, Lwt, and httpaf. The demo source code is available at the http-effects repo.
    • Sundries
      • ocaml-multicore/ocaml-multicore#406 Remove ephemeron usage of RPC

        The inter-domain mechanism is not required with the stop-the-world minor GC, and hence the same has been removed in the ephemeron implementation. The PR also does clean up and simplifies the ephemeron data structure and code.

      • ocaml-multicore/ocaml-multicore#411 Fix typo for presume and presume_arg in internal_variable_names

        A minor typo bug fix to rename Presume and Presume_arg in internal_variables_names.ml.

      • ocaml-multicore/ocaml-multicore#414 Fix up Ppoll semantics_of_primitives entry

        The semantics_of_primitives entry for Ppoll has been fixed which was causing flambda builds to remove poll points.

      • ocaml-multicore/ocaml-multicore#416 Fix callback effect bug

        The PR fixes a bug when the C-to-OCaml callback prevents effects crossing a C callback boundary. The stack parent is cleared before a callback, and restored afterwards. It also makes the stack parent a local root, so that the GC can see it inside the callback.


  • Ongoing
    • Configuration
      • ocaml-bench/ocaml-bench-scripts#12 Add support for parallel multibench targets and JSON input

        The RUN_CONFIG_JSON and BUILD_BENCH_TARGET variables are now added and passed during run-time for the execution of parallel benchmarks. Default values are specified so that the serial benchmarks can still run without explicitly requiring the same.

      • ocaml-bench/sandmark#180 Notebook Refactoring and User changes

        A refactoring effort is underway to make the parallel benchmark interactive. The user accounts on The Littlest JupyterHub installation have direct access to the benchmark results produced from ocaml-bench-scripts on the system.

      • ocaml-bench/sandmark#189 Add environment support for wrapper in JSON configuration file

        The OCAMLRUNPARAM is now passed as an environment variable to the benchmarks during runtime, so that, different parameter values can be used to obtain multiple results for comparison. The use case and the discussion are available at Running benchmarks with varying OCAMLRUNPARAM issue. The environment variables can be specified in the run_config.json file, as shown below:

           "name": "orun_2M",
           "environment": "OCAMLRUNPARAM='s=2M'",
           "command": "orun -o %{output} -- taskset --cpu-list 5 %{command}"
    • Proposals
      • ocaml-bench/sandmark#159 Implement a better way to describe tasklet cpulist

        The discussion to implement a better way to obtain the taskset list of cores for a benchmark run is still in progress. This is required to be able to specify hyper-threaded cores, NUMA zones, and the specific cores to use for the parallel benchmarks.

      • ocaml-bench/sandmark#179 [RFC] Classifying benchmarks based on running time

        A proposal to categorize the benchmarks based on their running time has been provided. The following classification types have been suggested:

        • lt_1s: Benchmarks that run for less than 1 second.
        • lt_10s: Benchmarks that run for at least 1 second, but, less than 10 seconds.
        • 10s_100s: Benchmarks that run for at least 10 seconds, but, less than 100 seconds.
        • gt_100s: Benchmarks that run for at least 100 seconds.

        The PR for the same is available at Classification of benchmarks.

      • We are exploring the use of opam-compiler switch environment to build the Sandmark benchmark test suite. The merge of systhreads compatibility support now enables us to install dune natively inside the switch environment, along with the other benchmarks. With this approach, we hope to modularize our benchmarking test suite, and converge to fully using the OCaml tools and ecosystem.
    • Sundries
      • ocaml-bench/sandmark#181 Lock-free map bench

        An implementation of a concurrent hash-array mapped trie that is lock-free, and is based on Prokopec, A. et. al. (2011). This cache-aware implementation benchmark is currently under review.

      • ocaml-bench/sandmark#183 Use crout_decomposition name for numerical analysis benchmark

        A couple of LU decomposition benchmarks exist in the Sandmark repository, and this PR renames the numerical-analysis/lu_decomposition.ml benchmark to crout_decomposition.ml. This is to address Rename lu_decomposition benchmark in numerical-analysis any naming confusion between the two benchmarks, as their implementations are different.

  • Completed
    • ocaml-bench/sandmark#177 Display raw baseline numbers in normalized graphs

      The raw baseline numbers are now included in the normalized graphs in the sequential notebook output. The graph for maxrsskb, for example, is shown below:


    • ocaml-bench/sandmark#178 Change to new Domain.DLS API with Initializer

      The multicore-minilight and multicore-numerical benchmarks have now been updated to use the new Domain.DLS API with initializer.

    • ocaml-bench/sandmark#185 Clean up existing effect benchmarks

      The PR ensures that the code compiles without any warnings, and adds a multicore_effects_run_config.json configuration file, and a run_all_effect.sh script to execute the same.

    • ocaml-bench/sandmark#186 Very simple effect microbenchmarks to cover code paths

      A set of four microbenchmarks to test the throughput of our effects system have now been added to the Sandmark test suite. These include effect_throughput_clone, effect_throughput_val, effect_throughput_perform, and effect_throughput_perform_drop.

    • ocaml-bench/sandmark#187 Implementation of 'recursion' benchmarks for effects

      A collection of recursion benchmarks to measure the overhead of effects are now included to Sandmark. This is inspired by the (Manticore benchmarks)[https://github.com/ManticoreProject/benchmark/].


  • Ongoing
    • ocaml/ocaml#9876 Do not cache young_limit in a processor register

      The PR removes the caching of young_limit in a register for ARM64, PowerPC and RISC-V ports. The Sandmark benchmarks are presently being tested on the respective hardware.

    • ocaml/ocaml#9934 Prefetching optimisations for sweeping

      The Sandmark benchmarking tests were performed for analysing a couple of patches that optimise sweep_slice, and for the use of prefetching. The objective is to reduce cache misses during GC.

  • Completed
    • ocaml/ocaml#9947 Add a naked pointers dynamic checker

      The check for "naked pointers" (dangerous out-of-heap pointers) is now done in run-time, and tests for the three modes: naked pointers, naked pointers and dynamic checker, and no naked pointers have been added in the PR.

    • ocaml/ocaml#9951 Ensure that the mark stack push optimisation handles naked pointers

      The PR adds a precise check on whether to push an object into the mark stack, to handle naked pointers.

    We would like to thank all the OCaml users and developers in the community for their continued support, reviews and contribution to the project.


  • AEIO: Asynchronous Effect-based IO
  • API: Application Programming Interface
  • ARM: Advanced RISC Machine
  • CPU: Central Processing Unit
  • DEC: Domain Execution Context
  • DLS: Domain Local Storage
  • GC: Garbage Collector
  • HTTP: Hypertext Transfer Protocol
  • JSON: JavaScript Object Notation
  • NUMA: Non-Uniform Memory Access
  • OPAM: OCaml Package Manager
  • OS: Operating System
  • PR: Pull Request
  • RISC-V: Reduced Instruction Set Computing - V
  • RPC: Remote Procedure Call
  • STW: Stop-The-World

Other OCaml News


If you happen to miss a CWN, you can send me a message and I'll mail it to you, or go take a look at the archive or the RSS feed of the archives.

If you also wish to receive it every week by mail, you may subscribe online.