Merge branch 'georges-tutorial' into 'dev'

Tutorial on the trace monad

See merge request ligolang/ligo!12
This commit is contained in:
Georges Dupéron 2019-06-12 00:27:46 +00:00
commit 91648d24b7
2 changed files with 212 additions and 28 deletions

View File

@ -1,6 +1,8 @@
(library (library
(name simple_utils) (name simple_utils)
(public_name simple-utils) (public_name simple-utils)
(preprocess
(pps simple-utils.ppx_let_generalized))
(libraries (libraries
yojson yojson
unix unix

View File

@ -1,3 +1,213 @@
(** Trace tutorial
The module below guides the reader through the writing of a
simplified version of the trace monad (`result`), and the
definition of a few operations that make it easier to work with
`result`.
*)
module Trace_tutorial = struct
(** The trace monad is fairly similar to the option type: *)
type 'a option =
Some of 'a (* Ok also stores a list of annotations *)
| None;; (* Errors also stores a list of messages *)
type annotation = string;;
type error = string;;
type 'a result =
Ok of 'a * annotation list
| Errors of error list;;
(** When applying a partial function on a result, it can return a valid result
(Some v), or indicate failure (None). *)
let divide a b =
if b = 0
then None
else Some (a/b);;
(** With the trace monad, the Errors case also indicates some information about
the failure, to ease debugging. *)
let divide_trace a b =
if b = 0
then (Errors [Printf.sprintf "division by zero: %d / %d" a b])
else Ok ((a/b) , []);;
(** when composing two functions, the error case is propagated. *)
let divide_three a b c =
let maybe_a_div_b = divide_trace a b in
match maybe_a_div_b with
Ok (a_div_b , _) -> divide_trace a_div_b c
| (Errors _) as e -> e;;
(** If both calls are successful, the lists of annotations are concatenated. *)
let divide_three_annots a b c =
let maybe_a_div_b = divide_trace a b in
match maybe_a_div_b with
Ok (a_div_b , annots1) ->
let maybe_a_div_b_div_c = divide_trace a_div_b c in
begin
match maybe_a_div_b_div_c with
Ok (a_div_b_div_c , annots2)
-> Ok (a_div_b_div_c , annots2 @ annots1)
| (Errors _) as e2 -> e2
end
| (Errors _) as e1 -> e1;;
(** This incurs quite a lot of noise, so we define a `bind` operator which
takes a function ('x -> ('y result)) and applies it to an existing
('x result).
* If the existing result is Errors, `bind` returns that error without
calling the function
* Otherwise `bind` unwraps the Ok and calls the function
* That function may itself return an error
* Otherwise `bind` combines the annotations and returns the second
result. *)
let bind f = function
| Ok (x, annotations) ->
(match f x with
Ok (x', annotations') -> Ok (x', annotations' @ annotations)
| Errors _ as e' -> ignore annotations; e')
| Errors _ as e -> e;;
(** The following function divide_three_bind is equivalent to the verbose
divide_three. *)
let divide_three_bind a b c =
let maybe_a_div_b = divide_trace a b in
let continuation a_div_b = divide_trace a_div_b c in
bind continuation maybe_a_div_b;;
(** This made the code shorter, but the reading order is a bit awkward.
We define an operator symbol for `bind`: *)
let (>>?) x f = bind f x;;
let divide_three_bind_symbol a b c =
let maybe_a_div_b = divide_trace a b in
let continuation a_div_b = divide_trace a_div_b c in
maybe_a_div_b >>? continuation;;
(** and we inline the two temporary let definitions: *)
let divide_three_bind_symbol' a b c =
divide_trace a b >>? (fun a_div_b -> divide_trace a_div_b c);;
(** This is now fairly legible, but chaining many such functions is
not the usual way of writing code. We use ppx_let to add some
syntactic sugar.
The ppx is enabled by adding the following lines inside the
section (library ) or (executable ) of the dune file for
the project that uses ppx_let.
(preprocess
(pps simple-utils.ppx_let_generalized))
*)
module Let_syntax = struct
let bind m ~f = m >>? f
module Open_on_rhs_bind = struct end
end;;
(** divide_three_bind_ppx_let is equivalent to divide_three_bind_symbol'.
Strictly speaking, the only difference is that the module
Open_on_rhs_bind is opened around the expression on the righ-hand side
of the `=` sign, namely `divide_trace a b` *)
let divide_three_bind_ppx_let a b c =
let%bind a_div_b = divide_trace a b in
divide_trace a_div_b c;;
(** This notation scales fairly well: *)
let divide_many_bind_ppx_let a b c d e f =
let x = a in
let%bind x = divide_trace x b in
let%bind x = divide_trace x c in
let%bind x = divide_trace x d in
let%bind x = divide_trace x e in
let%bind x = divide_trace x f in
Ok (x , []);;
(** We define a couple of shorthands for common use cases.
`ok` lifts a ('foo) value to a ('foo result): *)
let ok x = Ok (x, []);;
(** `map` lifts a regular ('foo -> 'bar) function on values
to a function on results, with type ('foo result -> 'bar result): *)
let map f = function
| Ok (x, annotations) -> Ok (f x, annotations)
| Errors _ as e -> e;;
(** `bind_list` turns a (('foo result) list) into a (('foo list) result).
If the list only contains Ok values, it strips the Ok returns that list
wrapped with Ok.
Otherwise, when one or more of the elements of the original list is
Errors, `bind_list` returns the first error in the list. *)
let rec bind_list = function
| [] -> ok []
| hd :: tl -> (
hd >>? fun hd ->
bind_list tl >>? fun tl ->
ok @@ hd :: tl
);;
(**
A major feature of Trace is that it enables having a stack of errors (that
should act as a simplified stack frame), rather than a unique error.
It is done by using the function `trace`.
For instance, let's say that you have a function that can trigger two errors,
and you want to pass their data along with an other error, what you would
usually do is:
```
let foobarer ... =
... in
let value =
try ( get key map )
with
| Bad_key _ -> raise (Foobar_error ("bad key" , key , map))
| Missing_value _ -> raise (Foobar_error ("missing index" , key , map))
in ...
```
With Trace, you would instead:
```
let foobarer ... =
... in
let%bind value =
trace (simple_error "error getting key") @@
get key map
in ...
```
And this will pass along the error triggered by "get key map".
*)
let trace err = function
| Ok _ as o -> o
| Errors errs -> Errors (err :: errs);;
(** The real trace monad is very similar to the one that we have
defined above. The main difference is that the errors and
annotations are structured data (instead of plain strings) and are
lazily-generated. *)
let the_end = "End of the tutorial.";;
end (* end Trace_tutorial. *)
module J = Yojson.Basic module J = Yojson.Basic
module JSON_string_utils = struct module JSON_string_utils = struct
@ -208,34 +418,6 @@ let internal_assertion_failure str = simple_error ("assertion failed: " ^ str)
*) *)
let dummy_fail = simple_fail "dummy" let dummy_fail = simple_fail "dummy"
(**
A major feature of Trace is that it enables having a stack of errors (that
should act as a simplified stack frame), rather than a unique error.
It is done by using the function `trace`.
For instance, let's say that you have a function that can trigger two errors,
and you want to pass their data along with an other error, what you would
usually do is:
```
let foobarer ... =
... in
let value =
try ( get key map )
with
| Bad_key _ -> raise (Foobar_error ("bad key" , key , map))
| Missing_value _ -> raise (Foobar_error ("missing index" , key , map))
in ...
```
With Trace, you would instead:
```
let foobarer ... =
... in
let%bind value =
trace (simple_error "error getting key") @@
get key map
in ...
```
And this will pass along the error triggered by "get key map".
*)
let trace info = function let trace info = function
| Ok _ as o -> o | Ok _ as o -> o
| Error err -> Error (fun () -> prepend_info (info ()) (err ())) | Error err -> Error (fun () -> prepend_info (info ()) (err ()))