resolve function

[jdidion]

One thing that has tripped up many people is concatenation of File and String. Specifically, it is not clear whether the following expressions should result in File or String:

• File + String
• String + File

I propose that we disallow all concatenations between File and String, and instead introduce a resolve function:

File resolve(String|File, String|File)

resolve accepts two arguments that are both assumed to be paths. The first argument may be a relative or absolute path; the second argument must be a relative path. resolve appends the two paths and returns the result as a File. Importantly, resolve ensures the two paths are joined by the system-specific path separator (i.e. / on *nix and \ on Windows).

[patmagee]

Just seeing this now. I think this is a great idea and definitely is inline with how files are treated in every other language. The implicit suggestion in wdl that "a file is just a special string" leads to soooo many issues and inconsistent behaviour. I would definitely +1 this proposal

[rhpvorderman]

Looks great! The way you describe it it brings Python's os.path.join to WDL. Which is very neat.

I would propose that this function can allow an arbitrary number of arguments, or an array. File resolve(String|File, String|File, ...) or File resolve(Array[String|File]). That would allow you to: File my_file = resolve('somewhere', 'over', 'the', 'rainbow') which yields somewhere/over/the/rainbow on *nix. Or if that is easier to implement for the execution engines File my_file = resolve(['somewhere', 'over', 'the', 'rainbow'])

My only gripe is with the name. resolving sounds like some 'thinking' is involved (i.e. ability to check if path exists, understanding .. etc.). I would prefer a name like pathjoin, which implies dumb, system-specific path separator aware behavior. Unless we of course want to understand . and ...
pathjoin also has my preference because the relation to filesystem paths is immediately clear, also to less programming oriented people.

[geoffjentry]

re what @rhpvorderman said, I was initially going to point out that this is effectively what java.nio.Path.resolve() does and thus we might be in a semantic debate based on people's biases from their daily driver programming language but that's not the case. In the JVM world this is semantically more like Paths.get() if I'm understanding correctly.

NB I don't think this should be called get 😄

[mr-c]

ensures the two paths are joined by the system-specific path separator (i.e. / on *nix and \ on Windows).

If OpenWDL has standardized on BASH, then implicitly you've standardized on POSIX, therefore the path separator should always be /, yes?