Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Change HIR visitor pattern #3160

Open
4 tasks
P-E-P opened this issue Sep 12, 2024 · 5 comments
Open
4 tasks

Change HIR visitor pattern #3160

P-E-P opened this issue Sep 12, 2024 · 5 comments
Assignees

Comments

@P-E-P
Copy link
Member

P-E-P commented Sep 12, 2024

HIR is in a bad state, multiple efforts could be started to improve the current situation. One of those could reduce the amount of code by removing boilerplate code within HIR visitors.

Most visitors currently have to redefine over and over traversal functions even when the visitor per se should only act on a few components.

Some proposals have been made to improve the general state of visitors within the project:

In the long run we should probably separate visit from walk in all our visitors but this requires an extensive refactor of multiple parts of the code base and multiple question remains (some AST visitors require infix operations).

We already provide an interface for HIR visitors: HIRFullVisitor.

  • Take a look at the AST visitor mechanisms (ASTDefaultVisitor, ASTVisitor, ASTContextualVisitor...)
  • Provide a default implementation for the HIR visitor.
  • Remove empty visits functions from HIR visitor implementors
  • Identify implementations in HIR visitor implementors where it could be replaced by a single call to parents' visit function.
@tamaroning
Copy link
Contributor

Hi, I recently encountered this issue and reached the same opinion that walk and visit should be separated.
#3115 (comment)

multiple question remains (some AST visitors require infix operations)

Do you mean mutating AST nodes by "infix" operations?

@P-E-P
Copy link
Member Author

P-E-P commented Sep 14, 2024

Hi, I recently encountered this issue and reached the same opinion that walk and visit should be separated. #3115 (comment)

multiple question remains (some AST visitors require infix operations)

Do you mean mutating AST nodes by "infix" operations?

We really need to be sure to where we're heading so I'll put a maximum of informations here from what has been discussed previously:

  • We need a default traversal operation ("walk")
    • to prevent huge file of copy/paste children visitor dispatch
    • or worse empty implementation
      • not visiting everything is faster but this means in the future modifying a child node might lead to unexpected results: the node won't be visited and recreating the traversal will be prone to errors
  • in ast: Check if the #[derive] attribute is used on a compatible item #2904 @GuillaumeGomez suggested that we separate walk and visit
    • this would clear up visit functions and separate mutations/actions done around the AST
      • I agree on those points but something is off: it puts some constraints on the traversal type
        • We could provide visit_prefix and visit_suffix instead of visit but I feel it is not "worth" the additional complexity
        • Not all traversals are prefix/suffix: some visitors make use of infix (rust ast collector), infix on N-tree means we cannot provide an easy interface
          • this edge case could be solved by providing a way to override "walk" (unexpected)
  • @CohenArthur would like in the long run to switch to a system with less side effects, every visit function should return it's own modifications, errors would be reported as error nodes instead of being collected alongside the traversal.
    • Incompatible with some visitors ? (Globbing visitor ?)
  • some bits are currently processed though visit but are not part of the visitor, it is incoherent

Our trees are quite big, this means any move will take some time to implement and will be tedious.

@CohenArthur
Copy link
Member

@CohenArthur would like in the long run to switch to a system with less side effects, every visit function should return it's own modifications, errors would be reported as error nodes instead of being collected alongside the traversal.

for what it's worth I'm not sure this is a good idea considering C++. this pattern works very well in Rust because move semantics are the default, and make sense. but it is a lot more annoying to do in C++

I really think that having separate walks and visits functions could clean things up, but it would be a huge change

@P-E-P
Copy link
Member Author

P-E-P commented Sep 17, 2024

For now we should probably focus on the initial goal of this PR: provide a default visitor, we'll then be able to move it to some walk funcitons in another issue/pr.

@Kamiinarii78
Copy link

@P-E-P Can you assign it to me ?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Status: To do
Development

No branches or pull requests

4 participants