Allow repr (+ str?) to take callables

[Julian]

It would be useful to be able to pass in a callable to do repr'ing of attributes.

For me this happens if I want to present a value in the repr differently than just repr(foo), which is sometimes useful, especially if the "different way" is to use e.g. reprlib.repr to do auto-truncation of long values.

Being able to pass in callables to the repr argument seems at first glance like a reasonably easy way to do that.

E.g.

attr.ib(repr=reprlib.repr)

and

attr.ib(repr=lambda value: repr(value) * 12)

You could preserve backwards compatibility by having True be equivalent to repr, and False be a sentinel (which you need anyways).

I think this could also satisfy #41, if you do the same to attr.s, and then provide an attr.simple_repr, so you'd do attr.s(repr=attr.simple_repr) and get the normal bracket repr instead of the eval one.

[blueyed]

See #224 (comment) for a patch to support a callable with repr.

[Tinche]

This is an interesting idea.

I think our repr support can be improved in any case; we're doing more work in the method that is generally needed, so it can be sped up by eval-ing a tailored function. Part of this work can be changing the Attribute.repr type to essentially be Union[bool, Callable[[], str] (so, either a boolean or a callable) and applying the callable when __repr__-ing.

[The-Compiler]

Another use case where this is useful are numbers which you'd like to show in hexadecimal (modifiers = attr.ib(repr=lambda mod: hex(int(mod))), or PyQt (not Python) enum values where you want to get a human-readable value instead of an int.

[nvgoldin]

Another use case I have is when you have a list attribute and you'd like __repr__ to print out only the list length (for example in cases where it is extremely large).

Would be happy to know if there is another way to do it.

[gimbo]

I've created a small extension which provides callable reprs and allows titles to be overridden too (e.g. for the length case cited above): attrs_flexible_reprs.py

Input/feedback welcomed.

[hynek]

@gimbo I think your flexible repr doesn’t take into account our recursion handler?

---

Re earlier comments, I wouldn’t make this issue depend on transforming our repr generator to code generation. It would be nice to have, but not required for fixing this particular issue.

[gimbo]

@hynek I'm sure you're right - it's very simplistic, just the simplest thing I came up with that scratched the itch I had.

I haven't delved into the code of attrs at all - when you say "recursion handler" I'm guessing you mean something to handle the case where a class A has a field also of type A? Certainly my code doesn't handle that gracefully, it seems. Perhaps I'll take a look at what attrs does by default and try to upgrade. :-) Thanks!

[hynek]

Glyph has recently taught our repr to detect loops in definitions so it doesn’t run into max recursion error. You'd have to look at the current code and shoehorn the callable thing into it.

[wbolster]

This is similar (identical?) to #567, and #568 adds support for this.

[hynek]

Took only two years!!!1!

[blueyed]

@hynek
Almost like good wine!