Skip to content

Commit

Permalink
Merge pull request #382 from panglesd/revamp-manual
Browse files Browse the repository at this point in the history
A more complete manual
  • Loading branch information
panglesd authored Jan 30, 2023
2 parents 4f4f7f7 + 35dbacf commit 1110af2
Show file tree
Hide file tree
Showing 34 changed files with 2,733 additions and 1,001 deletions.
114 changes: 114 additions & 0 deletions doc/ast-traversal.mld
Original file line number Diff line number Diff line change
@@ -0,0 +1,114 @@
{%html: <div style="display: flex; justify-content:space-between"><div>%}{{!"matching-code"}< Destructing AST nodes}{%html: </div><div>%}{{!"good-practices"}Good practices >}{%html: </div></div>%}

{0 AST Traversals}

The {{!Ppxlib.Parsetree}[Parsetree]} is a very complex type. Other {!Ppxlib} modules such as
{{!Ppxlib_metaquot}[Metaquot]}, {{!Ppxlib.Ast_builder}[Ast_builder]} and {{!Ppxlib.Ast_pattern}[Ast_pattern]} help in generating and matching values,
but only when the overall structure of the code is known in advance.

For other use cases, such as extracting all identifiers, checking that a
property is verified, or replacing all integer constants by something else,
those modules cannot really help. All these examples relate with another kind
of {{!Ppxlib.Parsetree}[Parsetree]} manipulations known as traversals.

A traversal is a recursive function that will be called on a value, and recursively on all
of its subvalues, combining the result in a certain way. For instance, {{!Stdlib.List.map}[List.map]} is a traversal of the
[list] type. In the case of a [list], a map is very simple to write, but in the
case of the long {{!Ppxlib.Parsetree}[Parsetree]} type, it is a lot of boilerplate code! Fortunately,
{{!Ppxlib}[ppxlib]} provides a way to ease this.

In [ppxlib], traversals are implemented using the "visitor" object-oriented pattern.

{1 Writing Traverses}

For each kind of traversal (described below), [ppxlib] provides a "default" traversal,
in the form of a class following the visitors pattern. For instance, in the case of the map traversal, the
default map is the identity AST map, and any object of class {{!Ppxlib.Ast_traverse.map}[Ast_traverse.map]}
will be this identity map. To apply a map to a node of a given type, one needs
to call the appropriate method:

{[
# let f payload =
let map = new Ppxlib.Ast_traverse.map in
map#payload ;;
val f : payload -> payload = <fun>
]}

In the example above, [f] is the identity map. But we want to define proper maps,
not just identity. This is done by creating a new class, making it inherit the
methods, and replacing the one that we want to replace. Here is an example, for
both the [iter] and [map] traversals:

{[
let f payload =
let checker =
object
inherit Ast_traverse.iter as super

method! extension ext =
match ext with
| { txt = "forbidden"; _ }, _ ->
failwith "Fordidden extension nodes are forbidden!"
| _ -> super#extension ext (* Continue traversing inside the node *)
end
in
let replace_constant =
object
inherit Ast_traverse.map
method! int i = i + 1
end
in
checker#payload payload;
replace_constant#payload payload
]}

Note that when redefining methods, unless explicitly wanting the traversal to
stop, the original method needs to be called! That should be all that’s necessary to
know and understand the {{!Ppxlib.Ast_traverse.map}API}.

{1 The Different Kinds of Traversals}

{{!Ppxlib}[ppxlib]} offers different kind of {{!Ppxlib.Parsetree}[Parsetree]} traversals:

- {{!Ppxlib.Ast_traverse.iter}Iterators}, which will traverse the type, calling
a function on each node for side effects.

- {{!Ppxlib.Ast_traverse.map}Maps}, where the content is replaced. A map will
transform a [Parsetree] into another [Parsetree], replacing nodes following the
map function.

- {{!Ppxlib.Ast_traverse.fold}Folds}, which will traverse the nodes, carrying a
value (often called an accumulator) that will be updated on each node.

- {{!Ppxlib.Ast_traverse.lift}Lifts}, a transformation that turns a [Parsetree] value in one of another type by
transforming it in a bottom-up manner. For instance, with a simple tree
structure, the corresponding [lift] function would be:

{[
let lift ~f = function
Leaf a -> f.leaf a
| Node(a,x,y) -> f.node a (lift ~f x) (lift ~f y)
]}

- Combinations of the two traversals, such as
{{!Ppxlib.Ast_traverse.fold_map}Fold-maps} and
{{!Ppxlib.Ast_traverse.lift_map_with_context}Lift-maps}.

- Variants of the above traversal, such as
{{!Ppxlib.Ast_traverse.map_with_context}Maps with context}, where a context
can be modified and passed down to child nodes during traversal. The context
never goes up; it is only propagated down. It is used for instance to track
opened module. To give a simple example, such a context could be the depth of
the current node, as in the following implementation for the simple tree type:

{[
let map_with_depth_context ~f ctxt = function
Leaf a -> f.leaf ctxt a
| Node(a,x,y) ->
f.node ctxt a
(map_with_depth_context (ctxt+1) ~f x)
(map_with_depth_context (ctxt+1) ~f y)
]}


{%html: <div style="display: flex; justify-content:space-between"><div>%}{{!"matching-code"}< Destructing AST nodes}{%html: </div><div>%}{{!"good-practices"}Good practices >}{%html: </div></div>%}
Loading

0 comments on commit 1110af2

Please sign in to comment.