🧐 Research a new solution for file management

[matthijseikelenboom]

Important

This issue should only be picked up by CodeEdit maintainers

As a CE developer
I want a reliable foundation that handles and manages files
Because it is crucial to other functionality of the application

Our current solution for file managements is reaching its limits and there are some serious bugs present. There has recently been an attempt at writing tests for this part of the code, but it revealed that there is unwanted functionality and results.

Acceptance criteria

• There has been a thorough investigation on how file management for CodeEdit can be done most effectively
• The findings have been documented

Scope

In scope

• Research options on how load a directory recursively as workspace
• Document the findings of the research (Can be a markdown file)
• Optionally make PoCs/Demos in a separate project

Out of scope

• Implement anything of the research done in CodeEdit

Requirements (Might need further refinement)

• The system should be able to load in a directory recursively from disk
• It has to be done asynchronously, so no problems occur when a folder is loaded with large amounts of content
• It should be able to monitor changes in all the files on disk inside that directory, this doesn't include the contents of the file
• It should be able to add a files and folders to disk

Risks

• Because CEWorkspaceFileManager and CEWorkspaceFile are referenced at so many places in the codebase, truly refactoring the code might prove difficult. Especially since Xcode's code analysis is plain bad.

Implementation design

Problems with current implementation

There are likely multiple ways to tackle this problem, in terms of using different Apple/Swift APIs and how we design the system ourselves. The latter is the most important and something that we can control. The current setup of CEWorkspaceFileManager and CEWorkspaceFile is not working properly. There are multiple cases where an "IllegalState" is achievable.

One of the reasons this occurs is because currently the CEWorkspaceFile contains all kinds of information, it contains multiple variables & functions related to Git, some that have todo with UI and some that have to do with tab related stuff (Which arguably is also UI). It's not clear what the responsibility of the class is. It doesn't look like the class isn't following the single responsibility principle anymore. Everything that is somewhat related to it, gets "dumped" in there.
It also has some properties that in theory have something to do with it, but from a system design point of view, they don't. An example is fileDocument, from the type CodeFileDocument. You can argue that it should be there because it represents the contents of the file and is therefore related to it. But it actually has more to do with the editor and truly modifying the contents of file. And CEWorkspaceFile represents the file in things like the project navigator, where the contents of the files don't matter.

Another thing that can (and should) be looked at is abstraction. The current CEWorkspaceFileManager and CEWorkspaceFile is referenced everywhere, and is therefore hard coupled to almost everything is the codebase. This also makes it more difficult to make unit tests, as the FileManager can not be mocked or stubbed. It also is a disadvantage in cases like these, be it requires changing all the files/components that rely on it.

Things that are interesting to look into when redesigning

Right now there is one CEWorkspaceFile, which can be either a Folder or a File. Perhaps this is the right way to do and it is how we want it, but another possibility is to make a distinction between the two by making separate classes.

If PoCs are made, it is advised to develop them test driven. Since this is core functionality of the application, we should have tests in place from the get go. This way we know how we want it to behave, before it has been coded.

Tasks

• Do research for one or multiple solutions (Multiple people can work on this, but should be in contact with each other to prevent duplicate work)
• Document the findings
• Discuss findings and PoC's with the team

[kdubb]

I propose a simple idea for an extensible solution: to use a node plus attachment model, much like the node tree hierarchy of most game engines.

A primary node would represent a file or folder with basic properties (e.g., size, creation date, modification date, etc.). Each node can then be extended (or decorated) with domain-specific data like change monitoring, git information, etc.

Swift's facilities, like generics, can be used to make this extended data easily accessible.

One downside of this approach is that the code requiring specific extended data must be guarded with existence checks, which can become a nuisance. This can be mitigated by throwing errors for non-existent data and using protocols that represent a "generic" node as one known to contain the required extended data.