OCaml Weekly News

Previous Week Up Next Week

Hello

Here is the latest OCaml Weekly News, for the week of September 08 to 15, 2020.

Table of Contents

A PPX Rewriter approach to ocaml-migrate-parsetree

Continuing this thread, Chet Murthy said

OK, I finished writing the migrations for the OCaml ASTs, so decided that it might be fun to write a little example to demonstrate how this works.

The ASTs

Suppose that I have two AST types (in ex_ast.ml):

module AST1 = struct
type t0 = string
type t1 = A of t1 * int list
type t2 = B of string * t0 | C of bool | D
type 'a pt3 = { it : 'a ; extra : int ; dropped_field: string }
type t4 = t2 pt3
end

module AST2 = struct
type t0 = int
type t1 = A of t1 * int list
type t2 = B of string * t0 | C of int | E
type 'a pt3 = { it : 'a ; extra : int ; new_field : int }
type t4 = t2 pt3
end

These AST types differ in the following ways:

  1. t0 is just completely different in these versions
  2. t1 is identical, but references a type that is different in each version; also an externally-defined polymorphic type-constructor (~ 'a list~)
  3. t2 differs in each version: it loses a branch (D), and gains a branch (E). It also has a branch whose arguments change type (C)
  4. 'a pt3 similarly loses a field and gains a field; it's also a polymorphic type
  5. t4 is identical, but again references types that are different in each version.

The Migrations

We'd like to write migrations that incur the -least- amount of boilerplate. What would that mean? For starters, we need to match up the types: there's no way to avoid that. Next, for branches that change type, we need to provide code. We need a way to flag branches that appear or disappear, and similarly for fields, to compute the value of fields that appear. So we'll:

  1. list the types we're going to work with, importing their definitions, so that the migration code will work against those already-defined types
  2. specify pairs of source-type, destination-type patterns, and for each, perhaps supply the function, or code for new/disappearing fields/branches
  3. for polymorphic types, we need to specify how their type-parameters get rewritten.

Below, we'll describe the process for taking this information and automatically computing the code of these migrations.

module Migrate_AST1_AST2 = struct

module SRC = Ex_ast.AST1
module DST = Ex_ast.AST2

exception Migration_error of string

let migration_error feature =
  raise (Migration_error feature)

let _migrate_list subrw0 __dt__ l =
  List.map (subrw0 __dt__) l

type t0 = [%import: Ex_ast.AST1.t0]
and t1 = [%import: Ex_ast.AST1.t1]
and t2 = [%import: Ex_ast.AST1.t2]
and 'a pt3 = [%import: 'a Ex_ast.AST1.pt3]
and t4 = [%import: Ex_ast.AST1.t4]
[@@deriving migrate
    { dispatch_type = dispatch_table_t
    ; dispatch_table_value = dt
    ; dispatchers = {
        migrate_list = {
          srctype = [%typ: 'a list]
        ; dsttype = [%typ: 'b list]
        ; code = _migrate_list
        ; subs = [ ([%typ: 'a], [%typ: 'b]) ]
        }
      ; migrate_t0 = {
          srctype = [%typ: t0]
        ; dsttype = [%typ: DST.t0]
        ; code = fun __dt__ s ->
            match int_of_string s with
              n -> n
            | exception Failure _ -> migration_error "t0"
        }
      ; migrate_t1 = {
          srctype = [%typ: t1]
        ; dsttype = [%typ: DST.t1]
        }
      ; migrate_t2 = {
          srctype = [%typ: t2]
        ; dsttype = [%typ: DST.t2]
        ; custom_branches_code = function
              C true -> C 1
            | C false -> C 0
            | D -> migration_error "t2:D"
        }
      ; migrate_pt3 = {
          srctype = [%typ: 'a pt3]
        ; dsttype = [%typ: 'b DST.pt3]
        ; subs = [ ([%typ: 'a], [%typ: 'b]) ]
        ; skip_fields = [ dropped_field ]
        ; custom_fields_code = {
            new_field = extra
          }
        }
      ; migrate_t4 = {
          srctype = [%typ: t4]
        ; dsttype = [%typ: DST.t4]
        }
      }
    }
]
end

module Migrate_AST2_AST1 = struct

module SRC = Ex_ast.AST2
module DST = Ex_ast.AST1

exception Migration_error of string

let migration_error feature =
  raise (Migration_error feature)

let _migrate_list subrw0 __dt__ l =
  List.map (subrw0 __dt__) l

type t0 = [%import: Ex_ast.AST2.t0]
and t1 = [%import: Ex_ast.AST2.t1]
and t2 = [%import: Ex_ast.AST2.t2]
and 'a pt3 = [%import: 'a Ex_ast.AST2.pt3]
and t4 = [%import: Ex_ast.AST2.t4]
[@@deriving migrate
    { dispatch_type = dispatch_table_t
    ; dispatch_table_value = dt
    ; dispatchers = {
        migrate_list = {
          srctype = [%typ: 'a list]
        ; dsttype = [%typ: 'b list]
        ; code = _migrate_list
        ; subs = [ ([%typ: 'a], [%typ: 'b]) ]
        }
      ; migrate_t0 = {
          srctype = [%typ: t0]
        ; dsttype = [%typ: DST.t0]
        ; code = fun __dt__ n -> string_of_int n
        }
      ; migrate_t1 = {
          srctype = [%typ: t1]
        ; dsttype = [%typ: DST.t1]
        }
      ; migrate_t2 = {
          srctype = [%typ: t2]
        ; dsttype = [%typ: DST.t2]
        ; custom_branches_code = function
              C 1 -> C true
            | C 0 -> C false
            | C _ -> migration_error "t2:C"
            | E -> migration_error "t2:E"
        }
      ; migrate_pt3 = {
          srctype = [%typ: 'a pt3]
        ; dsttype = [%typ: 'b DST.pt3]
        ; subs = [ ([%typ: 'a], [%typ: 'b]) ]
        ; skip_fields = [ new_field ]
        ; custom_fields_code = {
            dropped_field = string_of_int extra
          }
        }
      ; migrate_t4 = {
          srctype = [%typ: t4]
        ; dsttype = [%typ: DST.t4]
        }
      }
    }
]
end

Using the Migrations

And we can use them in the toplevel to migrate in each direction:

#load "ex_ast.cmo";;
#load "ex_migrate.cmo";;
open Ex_migrate ;;
open Ex_ast ;;
# Migrate_AST1_AST2.(dt.migrate_t4 dt AST1.{ it = C true ; extra = 3 ; dropped_field = "1" });;
- : Ex_migrate.Migrate_AST1_AST2.DST.t4 =
{Ex_migrate.Migrate_AST1_AST2.DST.it = Ex_migrate.Migrate_AST1_AST2.DST.C 1;
 extra = 3; new_field = 3}
# Migrate_AST2_AST1.(dt.migrate_t1 dt AST2.(A(1, [2;3])));;
- : Ex_migrate.Migrate_AST2_AST1.DST.t1 =
Ex_migrate.Migrate_AST2_AST1.DST.A ("1", [2; 3])

How Does It Work?

The migration code is computed in a pretty straightforward manner. In the following, assume we're migrating from AST1 to AST2 (DST is also the same as AST2). Let's call a "migration function" something of type

type ('a, 'b) migrater_t = dispatch_table_t -> 'a -> 'b

where dispatch_table_t will be defined later. First, assume (recursively) that each each migration rule yields a migration function. So a rule like

; migrate_t1 = {
    srctype = [%typ: t1]
  ; dsttype = [%typ: DST.t1]
  }

will yield a function of type

(AST1.t1, AST2.t1) migrater_t

and a rule like

; migrate_pt3 = {
    srctype = [%typ: 'a pt3]
  ; dsttype = [%typ: 'b DST.pt3]
  ; subs = [ ([%typ: 'a], [%typ: 'b]) ]
  ; skip_fields = [ dropped_field ]
  ; custom_fields_code = {
      new_field = extra
    }
  }

will yield

'a 'b. ('a, 'b) migrater_t -> ('a AST1.pt3, 'b AST2.pt3) migrater_t

Notice that this is a function that takes a migration from ~ 'a~ to ~ 'b~, and produces one from ~ 'a AST1.pt3~ to ~ 'a AST2.pt3~. So from a source-type, we need to mechanically compute the code, and also compute the result-type. This will be key idea: the migration process, applied to a source-type, will produce both the code that migrates that source-type, and also the destination-type.

Now, consider any rewrite rule, and let's argue by cases on how its code & destination-type can be computed.

  1. suppose that the source type (the field srctype) of the rule can be head-reduced (by applying some type-definition as an abbreviation) to an algebraic sum-type (A of ... | B of ...) or record-type ({a: t1 ; b : ... }). Then we can generate code to pattern-match on values of the source-type, and for variables in the generated patterns, we know their types. We can then apply (via pattern-matching over types) all the rewrite-rules to compute code that will migrate the variables' values. Since the recursive application of migration-rules produce destination-types, we can substitute those into branches/fields, to get the destination-type. [More on pattern-matching below.]
    • some of the branches of a sum-type might have their code specified in a custom_branches_code, and those branches supersede what would have been automatically-generated.
    • some of the fields of a record-type are listed in skip_fields, and they're not generated; some of the fields are listed with code to compute them (based on the fields in the pattern above) in custom_fields_code, and those get added.
  2. suppose that the source type after a single head-reduction yields a type-expression that can be successfully pattern-matched by one of the rewrite-rules other than this rule we're currently processing; then we can use that rewriter function to migrate the value of the source-type, and again, we get the destination-type.
    • Why other than the current rule? It's simple: we could end up in an infinite recursion if care isn't taken to write the migration rules correctly.
  3. The process of pattern-matching takes as input a type-expression, and a migration-rule. If the source-type of the migration-rule matches the type-expression (in the usual sense), then it produces bindings for the type-variables.

Let's look at an example. Suppose we want to compute the migration function for the type-expression t4.

  1. head-reducing once yields t2 pt3.
  2. the rule migrate_pt3 matches (source-type 'a pt3), with type-variable ~ 'a~ bound to t2.
    • also, (via the subs field) if the type bound to ~ 'a~ is rewritten to ~ 'b~ then the destination-type is ~ 'b AST2.pt3~ [remember this below]
  3. the type-expression t2 matches the rule migrate_t2, with no type-variables bound and destination-type AST2.t2.
  4. So migrate_t2 : (AST1.t2, AST2.t2) migrater_t (that is, the destination type is AST2.t2)
  5. Thus in #2 above, ~ 'b~ is bound to AST2.t2, and hence the destination-type in #2 above is AST2.t2 AST2.pt3

So the whole migration code for type t4 is migrate_pt3 migrate_t2 (which is what we would expect).

OCaml web server run multiple processes

mudrz announced

I've been running some simple benchmarks to compare web servers written in JS, F# and OCaml:

I started with @shonfeder's excellent tutorial: https://shonfeder.gitlab.io/ocaml_webapp/ Seriously if there were more tutorials like this the OCaml community would have been times bigger, they are so helpful when starting out and you don't want to solve the same problem everyone else has already solved. And then implemented parts of it in JS and F#

the test:

  • queries a DB
  • adds an additional item to the result
  • orders the list
  • returns an html containing the rendered result

I ran each test 3 times and posted the best result

This is not a scientific test using a controlled environment, this is a quick and dirty test, so take it with a bit of salt (but in many ways the results are not so unexpected). The tests are also not comparing the languages themselves but a combination of some of the more popular libraries of them, which is how you would normally use them.

I'm not posting a link to a repo with the benchmarks since the code is not organised and the tests are not automated, don't want to spend too much time on this, but happy to post the code used if someone is interested.

The tests were made with wrk -t8 -c400 -d30s - 8 threads, 400 connections, 30 seconds;

HTML endpoint, no DB access

JS - single process

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    44.16ms    4.73ms  59.22ms   88.44%
    Req/Sec     1.10k    89.00     1.50k    82.83%
  262651 requests in 30.05s, 179.60MB read
  Socket errors: connect 0, read 264, write 0, timeout 0
Requests/sec:   8739.98
Transfer/sec:      5.98MB

JS - multiple processes

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    43.37ms   56.91ms 316.27ms   83.49%
    Req/Sec     2.43k   459.80     3.81k    73.25%
  580742 requests in 30.05s, 397.10MB read
  Socket errors: connect 0, read 239, write 0, timeout 0
Requests/sec:  19325.95
Transfer/sec:     13.21MB

OCaml - single process

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    15.01ms    2.28ms  39.86ms   89.00%
    Req/Sec     3.27k   240.86     4.53k    89.83%
  781313 requests in 30.01s, 489.54MB read
  Socket errors: connect 0, read 308, write 0, timeout 0
Requests/sec:  26031.65
Transfer/sec:     16.31MB

F#

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     5.46ms  840.10us  20.34ms   83.20%
    Req/Sec     8.91k   659.89    13.27k    73.25%
  2130565 requests in 30.04s, 1.45GB read
  Socket errors: connect 0, read 272, write 0, timeout 0
Requests/sec:  70923.15
Transfer/sec:     49.31MB

Results:

  • OCaml: 1.34x more requests/s than JS, with 2.8x less latency
  • F#: 3.66x more than JS, 2.7x more requests/s than OCaml, with 2.7x less latency

JSON endpoint, DB access

The JSON response for JS and OCaml was:

{"excerpts":[{"author":"kan","excerpt":"Another excerpt","source":"My source2","page":"another
page"},{"author":"kan","excerpt":"My excerpt","source":"my source","page":"23"}]}

for F# it was slightly longer since option types are serialized by default with a Some/None variant (it can be changed):

{"excerpts":[{"author":"kan","excerpt":"Another excerpt","source":"My
source2","page":{"case":"Some","fields":["another page"]}},{"author":"kan","excerpt":"My
excerpt","source":"my
source","page":{"case":"Some","fields":["23"]}},{"author":"a","excerpt":"b","source":"c","page":{"case":"Some","fields":["d"]}}]}

The DB was Postgres with 10 max connections

JS - single process

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    57.79ms    5.68ms  88.02ms   85.24%
    Req/Sec   848.20    109.95     1.37k    72.24%
  202885 requests in 30.09s, 62.69MB read
  Socket errors: connect 0, read 237, write 0, timeout 0
Requests/sec:   6742.09
Transfer/sec:      2.08MB

JS - multiple processes

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    57.48ms   61.84ms 774.03ms   83.36%
    Req/Sec     1.20k   260.84     2.29k    71.38%
  287101 requests in 30.04s, 88.71MB read
  Socket errors: connect 0, read 286, write 38, timeout 0
Requests/sec:   9558.04
Transfer/sec:      2.95MB

OCaml

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     6.69ms   38.78ms   1.07s    98.17%
    Req/Sec     1.78k   842.50     3.62k    56.33%
  424454 requests in 30.02s, 100.39MB read
  Socket errors: connect 0, read 253, write 0, timeout 13
Requests/sec:  14139.42
Transfer/sec:      3.34MB

F#

  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    19.27ms    3.92ms 107.53ms   82.60%
    Req/Sec     2.54k   165.82     3.26k    79.21%
  606868 requests in 30.02s, 261.02MB read
  Socket errors: connect 0, read 259, write 0, timeout 0
Requests/sec:  20214.71
Transfer/sec:      8.69MB

Results:

  • OCaml: 1.48x more requests/s than JS (up from 1.34x before), with 8.6x less latency (before: 2.7x)
  • F#: 2.1x more than JS (down from 3.66x before), 1.43x more requests/s than OCaml (down from 2.7x before), with 2.88x MORE latency than OCaml (before: 2.7x LESS than OCaml)

Observations:

  • JS is performing unexpectedly good compared to compiled languages
  • F# (or ASP.NET Core) is really fast out of the box, with no tweaking necessary
  • OCaml is running on a single process and has had Max request time of 1.07s and Stdev 10x that of F#; in some tests it spiked to 2seconds for some requests, is this the GC? how can I troubleshoot that?

Is there a good tutorial on running OCaml with multiple processes and generally commonly faced use cases for web servers? There are countless articles for the other ecosystems, but it is a bit difficult to find ones for OCaml, making it a bit time consuming to try to figure each thing out

Later, mudrz added

added the code to: https://gitlab.com/mudrz/ocaml-web-benchmarks/-/tree/master/

there are 3 dirs for each server

mudrz also said

I found it generally interesting to compare OCaml and F# syntaxes and how things work, so adding this if others find it interesting:

HTML templates

OCaml (tyxml): https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/opi-bench/lib/content.ml#L5-42 each element is a function that has labeled arguments for attributes and a final argument for the children

head
  (title (txt "OCaml Webapp Tutorial"))
  [ meta ~a:[a_charset "UTF-8"] ()
  ; link ~rel:[`Stylesheet] ~href:"/static/style.css" () ]

F# (giraffe viewing engine, there were others, but I haven't tested them): https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/fsharp-bench/src/fsharp-bench/Content.fs#L6-35 each element is a function that accepts 2 lists - the first one for attributes, the second for children

head [] [
    title []  [ encodedText "fsharp_bench" ]
    link [ _rel  "stylesheet"
	   _type "text/css"
	   _href "/style.css" ]
]

DB Query

OCaml (caqti rapper): https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/opi-bench/lib/db.ml#L78-88 using a ppx

[%rapper get_many
    {sql|
    SELECT @string{author}, @string{excerpt}, @string{source}, @string?{page}
    FROM excerpts
    WHERE author = %string{author}
    |sql}
    record_out
]

F# (dapper): https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/fsharp-bench/src/fsharp-bench/Db.fs#L30-35

let sql = """
   SELECT author, excerpt, source, page
   FROM excerpts
   WHERE author = @author;
"""
let! data = conn.QueryAsync<Excerpt.t>(sql,  dict ["author" => "kan"])

Serialization

OCaml: define as serialisable: https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/opi-bench/lib/excerpt.ml#L1-6

type t =
  { author: string
  ; excerpt: string
  ; source: string
  ; page: string option
  }[@@deriving yojson]

and then call the generated function: https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/opi-bench/lib/route.ml#L97

let open Excerpt.Response.Err in
let json = to_yojson { message= e } in

F#: mark type as "cli mutable": https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/fsharp-bench/src/fsharp-bench/Excerpt.fs#L3-9

[<CLIMutable>]
type t =
  { author: string
  ; excerpt: string
  ; source: string
  ; page: string option
  }

serialisation happens automatically

async:

OCaml (lwt): https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/opi-bench/lib/route.ml#L106-110 "await" with let* and let+ and then return with Lwt.return

let excerpts = get "/excerpts" begin fun req ->
    let open Lwt.Syntax in
    let+ authors = Db.Get.authors req in
    respond_or_err Content.author_excerpts_page authors
  end

F#: https://gitlab.com/mudrz/ocaml-web-benchmarks/-/blob/master/fsharp-bench/src/fsharp-bench/Program.fs#L81-86

fun ctx next -> task {
    let! excerpts = Db.fortunes ()
    let res: Db.Excerpt.res = { excerpts= excerpts }
    return! json res ctx next
}

Query-json: Re-implemented jq in Reason Native/OCaml

David Sancho announced

I re-implementatied jq in Reason Native and OCaml and I'm a little proud :star:, started as a way to learn how to write lexers/parsers/compilers with the OCaml stack, but ending up very functional.

I'm using it right now, every day. It's called query-json or for short "q".

https://github.com/davesnx/query-json

It has better performance (between 2x and 5x) than jq, richer error messages and simplified API. It's mostly thanks to OCaml/Reason, rather than my skill to code a compiler! Still isn't feature complete, though, but have implemented most of the common functionality, Adding those features shouldn't affect the performance, so I will keep adding them with time.

You can check the benchmarks here: https://github.com/davesnx/query-json#Performance

The main idea is to improve the api of the operations and errors with manipulation JSON files.

If you don't know what jq is, check thoughtbot dot com/blog/jq-is-sed-for-json or programminghistorian dot org/en/lessons/json-and-jq.

Hope you like it and let me know if there's something that you struggled with jq and I can make it better in q, let me know, I'm always open to any DM.

Thanks 👋🏽

PS: The creator of jq (@stedolan), which is a current OCaml core maintainer is doing ocaml-multicore, which at some point could improve the "q" performance even more.

Suggestions for a simple, portable, graphics library?

Yaron Minsky asked

I'm working on a little side project where I need to do some simple, cross-platform game-like graphics. Really simple: I want to be able to draw some colored lines, and splat out rotated images. But I'd also like this to be reasonably efficient, and reasonable easy to use.

And, ideally, I'd love for something that was portable to Windows.

I've done just a bit of poking around, and it wasn't quite clear to me what the right choice was.

  • I looked at tsdl and tgls, but it wasn't clear to me with either of them how to rotate a texture. Maybe I'm just missing something? Also, I found tgls a little hard to navigate, since I haven't used opengl for 20 years. I feel like rotation has to be in there, but searching for "rotate" in the docs comes up empty, and I couldn't find any good examples. The portability story here seemed hopeful.
  • I looked at OCaml's venerable Graphics library. Super easy to use, not sure about the portability story.
  • wall. Seems to have all the capabilities, not sure about the portability story.
  • web. Portable, though I'd need to have a more complex setup for users, where the native OCaml program would run a web-server that would communicate to the OCaml UI, since I need this to be a native app. Feels extra complicated.

Anyway, does anyone have any suggestions on what path to take? Am I missing any options?

Daniel Bünzli replied

I looked at tsdl and tgls , but it wasn’t clear to me with either of them how to rotate a texture. Maybe I’m just missing something?

I have never really used the SDL 2D support but it could have you covered. This should allow you to rotate a texture.

Maybe I’m just missing something? Also, I found tgls a little hard to navigate, since I haven’t used opengl for 20 years. I feel like rotation has to be in there, but searching for “rotate” in the docs comes up empty, and I couldn’t find any good examples. The portability story here seemed hopeful.

The OpenGL of today bares little ressemblance to the one you knew, this is the hello world nowadays. Rotation is there under the form of matrices that you access in shaders but you also need to deal with defining, compiling and feeding those with data which brings quite a bit of boiler plate (see the example).

Also on macOS, OpenGL "the API", seems to be on the way out. I suspect the 2D SDL API may be a better bet since it will likely keep you with a compatibility, hardware accelerated layer.

If you can lift the native constraint, the web would still be my first shot, nothing beats it's ease of installation.

holmdunc added

Yeah, SDL already uses the Metal rendering backend by default on macOS. Can be confirmed using this function.

Tobias Mock also replied

You surely can't go wrong with SDL/tsdl.

Another alternative could be raylib, a gamedev library created mainly for teaching. The lib has tons of examples, of which a few are already translated to the OCaml bindings. There is also an example for rotating a texture. And it works on Windows.

Ohad Rau said

Looks like I'm a little late to the thread but I think reason-skia is an awesome choice. It should provide all the Skia primitives which are pretty similar to the JavaScript canvas API and it's specifically focused on 2d graphics. It's also got great cross-platform support. It's being very actively developed as part of Revery, so it seems much more up to date than tsdl/tgls (IIRC Bryan, the maintainer for Revery, developed his own SDL bindings since tsdl was too out of date).

It's also possible to use the Skia bindings through Revery to take advantage of the rest of Revery's features: Revery exposes Skia as part of its Canvas API. This is really nice because you can avoid all the window setup stuff you'd have to do with SDL or raw OpenGL.

4af4285b00a0712cc751c703932dbcbad3e5800d.png

Tom Ekander added

If you excuse the Reason-syntax, here's a "Hello World"-example for Revery using Canvas.

0d8e788aa1770f2462af2aad7ddce98f9a0201e0_2_1035x739.jpeg

Daniel Bünzli then asked

it seems much more up to date than tsdl/tgls (IIRC Bryan, the maintainer for Revery, developed his own SDL bindings since tsdl was too out of date).

Before false information starts to spread could you please make precise what is "too out of date" in tsdl ?

A few things that made it in SDL 2.0.7-11 may not be in (but some of these things were actually contributed, I don't remember exactly which ones).

But given the surface of the API and as you can witness by reading the linked release notes if something is missing that's not much and rather trivial to add.

1sr release of omlr: Multiple Linear Regression modeling (using R under the carpet)

UnixJunkie announced

omlr is now available in opam.

# opam info omlr

<><> omlr: information on all versions ><><><><><><><><><><><><><><><><><><><><>
name                   omlr
all-installed-versions 1.0.2 [default]
all-versions           1.0.2

<><> Version-specific details <><><><><><><><><><><><><><><><><><><><><><><><><>
version       1.0.2
repository    github
url.src:      "https://github.com/UnixJunkie/omlr/archive/v1.0.2.tar.gz"
url.checksum: "md5=4b637863b4bc4ed2e27de45cf0872736"
homepage:     "https://github.com/UnixJunkie/omlr"
bug-reports:  "https://github.com/UnixJunkie/omlr/issues"
dev-repo:     "git+https://github.com/UnixJunkie/omlr.git"
authors:      "Francois Berenger"
maintainer:   "unixjunkie@sdf.org"
license:      "BSD-3"
depends:      "batteries"
              "conf-gnuplot"
              "conf-r"
              "cpm"
              "dolog" {>= "4.0.0"}
              "dune" {>= "1.11"}
              "minicli" {>= "5.0.0"}
              "ocaml"
synopsis      Multiple Linear Regression model
description   Train a MLR model using R.
              usage:
              ./mlr_model
                [-i <input.csv>]: input CSV file
                [--NxCV <int>]: number of folds of cross validation
                [-s|--save <filename>]: save model to file
                [-l|--load <filename>]: restore model from file
                [-o <filename>]: predictions output file
                [--no-shuffle]: do not randomize input lines
                [--no-header]: CSV file has no header
                [--no-plot]: don't call gnuplot
                [-d <char>]: field delimited in CSV file (default=',')
                [-v]: verbose/debug mode
                [-h|--help]: show this message

The project is here: https://github.com/UnixJunkie/omlr

Note that it could be entirely programmed using owl. However, I find R more portable than owl, hence the hack.

Multicore OCaml: August 2020

Anil Madhavapeddy announced

Welcome to the August 2020 Multicore OCaml report (a few weeks late due to August slowdown). This update along with the previous updates have been compiled by @shakthimaan, @kayceesrk and myself.

There are some talks related to multicore OCaml which are now freely available online:

The Multicore OCaml project has had a number of optimisations and performance improvements in the month of August 2020:

  • The PR on the implementation of systhreads with pthreads continues to undergo review and improvement. When merged, this opens up the possibility of installing dune and other packages with Multicore OCaml.
  • Implementations of mutex and condition variables is also now under review for the `Domain` module.
  • Work has begun on implementing GC safe points to ensure reliable, low-latency garbage collection can occur.

We would like to particularly thank these external contributors:

  • Albin Coquereau and Guillaume Bury for their comments and recommendations on building Alt-Ergo.2.3.2 with dune.2.6.0 and Multicore OCaml 4.10.0 in a sandbox environment.
  • @Leonidas for testing the code size metric implementation with Core and Async, and for code review changes.

Contributions such as the above towards adapting your projects with our benchmarking suites are always most welcome. As with previous updates, we begin with the Multicore OCaml updates, which are then followed by the enchancements and bug fixes to the Sandmark benchmarking project. The upstream OCaml ongoing and completed tasks are finally mentioned for your reference.

Multicore OCaml

  • Ongoing
    • ocaml-multicore/ocaml-multicore#381 Reimplementating systhreads with pthreads

      This PR has made tremendous progress with additions to domain API, changes in interaction with the backup thread, and bug fixes. We are now able to build dune.2.6.1 and utop with this PR for Multicore OCaml, and it is ready for review!

    • ocaml-multicore/ocaml-multicore#384 Add a primitive to insert nop instruction

      The nop primitive is introduced to identify the start and end of an instruction sequence to aid in debugging low-level code.

    • ocaml-multicore/ocaml-multicore#390 Initial implementation of Mutexes and Condition Variables

      A draft proposal that adds support for Mutex variables and Condition operations for the Multicore runtime.

  • Completed
    • Optimisations
      • ocaml-multicore/domainslib#16 Improvement of parallel_for implementation

        A divide-and-conquer scheme is introduced to distribute work in parallel_for, and the chunk_size is made a parameter to improve scaling with more than 8-16 cores. The blue line in the following illustration shows the improvement for few benchmarks in Sandmark using the default chunk_size along with this PR: d2cb71fa73825cce9231538e213d1815dde0f139_2_697x750.png

      • ocaml-multicore/multicore-opam Use -j%{jobs}% for multicore variant builds

        The use of -j%{jobs}% in the build step for multicore variants will speed up opam installs.

      • ocaml-multicore/ocaml-multicore#374 Force major slice on minor collection

        A minor collection will need to schedule a major collection, if a blocked thread may not progress the major GC when servicing the minor collector through handle_interrupt.

      • ocaml-multicore/ocaml-multicore#378 Hold onto empty pools if swept while allocating

        An optimization to improve pause times and reduce the number of locks by using a release_to_global_pool flag in pool_sweep function that continues to hold onto the empty pools.

      • ocaml-multicore/ocaml-multicore#379 Interruptible mark and sweep

        The mark and sweep work is now made interruptible so that domains can enter the stop-the-world minor collections even if one domain is performing a large task. For example, for the binary tree benchmark with four domains, major work (pink) in domain three stalls progress for other domains as observed in the eventlog. ef309b229845e21ce8d2753ef30009139e6da5f0_2_808x750.png

        With this patch, we can observe that the major work in domains two and four make progress in the following illustration: 1b731fa02e721c6b00a5a81080a7517742751aab_2_982x750.png

      • ocaml-multicore/ocaml-multicore#380 Make DLS call to caml_domain_dls_get @@noalloc

        The caml_dls_get is tagged with @@noalloc to reduce the C call overhead.

      • ocaml-multicore/ocaml-multicore#382 Optimise caml_continuation_use_function

        A couple of optimisations that yield 25% performance improvements for the generator example by using caml_domain_alone, and using caml_gc_log under DEBUG mode.

      • ocaml-multicore/ocaml-multicore#389 Avoid holding domain_lock when using backup thread

        The wait time for the main OCaml thread is reduced by altering the backup thread logic without holding the domain_lock for the BT_IN_BLOCKING_SECTION.

    • Sundries

Benchmarking

  • Ongoing
    • ocaml-bench/sandmark#122 Measurements of code size

      The code size of a benchmark is one measurement that is required for flambda branch. A PR has been created that now emits a count of the CAML symbols in the output of a bench result as shown below:

      {"name":"knucleotide.", ... ,"codesize":276859.0, ...}
      
    • ocaml-bench/sandmark#169 Add check_url for .json and pkg-config, m4 in Makefile

      A check_url target in the Makefile has been defined to ensure that the ocaml-versions/*.json files have a URL parameter. The patch also adds pkg-config and m4 to Ubuntu dependencies.

  • Completed
    • Benchmarks
      • ocaml-bench/sandmark#107 Add Coq benchmarks

        The fraplib library from the Formal Reasoning About Programs has been dunified and included in Sandmark for Coq benchmarks.

      • ocaml-bench/sandmark#151 Evolutionary algorithm parallel benchmark

        The evolutionary algorithm parallel benchmark is now added to Sandmark.

      • ocaml-bench/sandmark#152 LU decomposition: random numbers initialisation in parallel

        The random number initialisation for the LU decomposition benchmark now has parallelism that uses Domain.DLS and Random.State.

      • ocaml-bench/sandmark#153 Add computationally intensive Coq benchmarks

        The BasicSyntax and AbstractInterpretation Coq files perform a lot of minor GCs and allocations, and have been added as benchmarks to Sandmark.

      • ocaml-bench/sandmark#155 Sequential version of Evolutionary Algorithm

        The sequential version of algorithms are used for comparison with their respective parallel implementations. A sequential implementation for the Evolutionary Algorithm has now been included in Sandmark.

      • ocaml-bench/sandmark#157 Minilight Multicore: Port to Task API and DLS for Random States

        The Minilight benchmark has been ported to use the Task API along with the use of Domain Local Storage for the Random States. The speedup is shown in the following illustration:

        89261588-f5072980-d64b-11ea-9364-3f18342b7274.png

      • ocaml-bench/sandmark#164 Tweaks to multicore-numerical/game_of_life

        The board_size for the Game of Life numerical benchmark is now configurable, and can be supplied as an argument.

    • Bug Fixes
      • ocaml-bench/sandmark#156 Fix calculation of Nbody Multicore

        Minor fixes in the calculation of interactions of the bodies in the Nbody implementation, and use of local ref vars to reduce writes and cache traffic.

      • ocaml-bench/sandmark#158 Fix key error for Grammatrix for Jupyter notebook

        The Key Error issue with notebooks/parallel/parallel.ipynb is now resolved by passing a value to params in the multicore_parallel_run_config.json file.

    • Sundries
      • ocaml-bench/sandmark#154 Revert PARAMWRAPPER changes

        Undo the PARAMWRAPPER configuration for parallel benchmark runs in the Makefile, as they are not required for sequential execution.

      • ocaml-bench/sandmark#160 Specify prefix,libdir for alt-ergo sandbox builds

        The alt-ergo library and parser require the prefix and libdir to be specified with configure in order to build in a sandbox environment. The initial discussion is available at OCamlPro/alt-ergo#351.

      • ocaml-bench/sandmark#162 Avoid installing packages which are unused for Multicore runs

        The PACKAGES variable in the Makefile has been simplified to include only those dependency packages that are required to build Sandmark.

      • ocaml-bench/sandmark#163 Update to domainslib 0.2.2 and use default chunk_size

        The domainslib dependency package has been updated to use the 0.2.2 released version, and chunk_size for various benchmarks uses num_tasks/num_domains as default.

OCaml

  • Ongoing
    • ocaml/ocaml#9756 Garbage collectors colour change

      The PR is needed for use with the Multicore OCaml major collector by removing the need of gray colour in the garbage collector (GC) colour scheme.

  • Completed
    • ocaml/ocaml#9722 EINTR-based signals, again

      The patch provides a new implementation to solve a collection of locking, signal-handling and error checking issues.

    Our thanks to all the OCaml developers and users in the community for their support and contribution to the project. Stay safe!

Acronyms

  • API: Application Programming Interface
  • DLS: Domain Local Storage
  • GC: Garbage Collector
  • OPAM: OCaml Package Manager
  • LU: Lower Upper (decomposition)
  • PR: Pull Request

Old CWN

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.