Remove non-`_mut` function variants. about imageproc

[ripytide]

This came up when working on #602 where the non-mut functions double the amount of work I would have to do.

[theotherphil]

I expect the non-mut versions will be far more widely used, with the _mut versions existing solely to allow users to control allocations if this matter for their use cases. I'm not a fan of removing the more ergonomic and more used (I assume) versions of functions.

We should standardise on having detailed documentation on the non-mut versions and linking to this from the _mut versions.

In theory the generation of the _mut versions could be automated via a proc macro but this feels unlikely to be worth the extra complexity or compile time.

[ripytide]

Judging by some grep.app searches it varies quite a bit.

dilate shows only the mut version is used.
map_pixels shows both the mut version and the non-mut version.

[ripytide]

In theory the generation of the _mut versions could be automated via a proc macro but this feels unlikely to be worth the extra complexity or compile time.

I think a proc macro would solve my issues with maintainability and help with standardization, I don't think the added compile time would be noticeable for such a simple proc-macro, the complexity would be mostly a one-time cost on the implementation of the proc-macro and should be very easy to use after its created perhaps?

[ripytide]

Nevermind, I tried implementing the proc-macro and I needed to use a lot of heuristics when generating the non-mut function signature such as choosing a return type based on the input image type which could get very complicated very quickly when generics are involved.

What about switching the documentation to the mut version rather than the non-mut function seeing as that is the "real" function whereas the non-mut variation is simply a useful helper function? Then we could standardize the documentation of the non-mut functions as:

/// An immutable version of [`erode_mut`].

This function clones the given image, then runs the operation on the cloned image and then returns the mutated image.

This documentation could be generated from a regular macro_rules!(erode_mut) as well which reduces the amount of duplication.

[theotherphil]

It’s more convenient to write doctests for the non-mutating variants, so I’d flip your suggestion. But the idea of generating the _mut docs via a macro is a good one.

[ripytide]

But is it a good idea to be teaching new users to use the less-efficient functions? That may cause users to use the non-mut variations even if they don't need the original image anymore since those are the functions with examples. That doesn't really fit with rusts fast-by-default design imo.

[theotherphil]

The two versions of each function will be next to each other in the documentation so anyone who wants to re-use an allocation can quickly spot that this is possible.

[ripytide]

Fair point, I'll start a PR standardizing all the mut/non-mut functions using a macro then 👍 After that, I'll be able to rebase and continue #602 which might become easier.

[theotherphil]

Sounds good, thanks.