Allow slf: PyRef<Self>/PyRefMut<Self> in pymethods

[kngwyu]

Looks like

#[pymethods]
impl Reader {
    fn get_iter(slf: PyRef<Self>, keys: Py<PyBytes>) -> PyResult<Iter> {
        Ok(Iter {
            reader: slf.into(),
            keys: keys,
            idx: 0,
        })
    }
}

and can be called like usual methods from python.
Resolves #356 (comment)
Currently only slf: PyRef is tested.
Another concern is I observed an undesirable segfault when using arg: PyRef<PyBytes> in pymethod(though I'm not sure it's related to this PR

[konstin]

In general for the self type I think it's better if we had a trait (or maybe reuse extract) that is implemented for all three types instead of trying to guess the right method based on the string-representation of the type

[codecov]

Codecov Report

Merging #419 into master will decrease coverage by 0.06%.
The diff coverage is 96.29%.

@@            Coverage Diff             @@
##           master     #419      +/-   ##
==========================================
- Coverage   88.06%   87.99%   -0.07%     
==========================================
  Files          83       65      -18     
  Lines        3593     3314     -279     
==========================================
- Hits         3164     2916     -248     
+ Misses        429      398      -31

Impacted Files	Coverage Δ
src/lib.rs	0% <ø> (ø)	⬆️
src/instance.rs	96.77% <100%> (+0.32%)	⬆️
src/python.rs	93.58% <100%> (+2%)	⬆️
src/conversion.rs	96.46% <93.93%> (-1.04%)	⬇️
src/class/number.rs	94.2% <0%> (-0.82%)	⬇️
src/type_object.rs	88.4% <0%> (-0.64%)	⬇️
src/class/sequence.rs	96.42% <0%> (-0.59%)	⬇️
src/class/buffer.rs	100% <0%> (ø)	⬆️
src/ffi2/setobject.rs
... and 22 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bc7776e...031f14e. Read the comment docs.

[codecov]

Codecov Report

Merging #419 into master will increase coverage by 0.16%.
The diff coverage is 96.49%.

@@            Coverage Diff             @@
##           master     #419      +/-   ##
==========================================
+ Coverage   87.66%   87.82%   +0.16%     
==========================================
  Files          63       64       +1     
  Lines        3355     3385      +30     
==========================================
+ Hits         2941     2973      +32     
+ Misses        414      412       -2

Impacted Files	Coverage Δ
src/lib.rs	0% <ø> (ø)	⬆️
src/python.rs	93.58% <100%> (+2%)	⬆️
src/instance.rs	96.83% <100%> (+0.38%)	⬆️
src/conversion.rs	96.46% <93.93%> (-1.04%)	⬇️
src/ffi3/bytesobject.rs	100% <0%> (ø)
src/types/string.rs	89.83% <0%> (+1.15%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f8bf258...c7d6c48. Read the comment docs.

[kngwyu]

I think it's better if we had a trait

Experimentally I introduced a new trait named FromPyPointer, which registers the pointer to GIL pool and returns the reference of it.
If you come up with a better design, please let me know.

[konstin]

I'll have to check if we already have the traits that can do what FromPyPointer does or with we can replace some other code with the new FromPyPointer, but otherwise it's looking good

[kngwyu]

@konstin
I want this PR move forward.
Though FromPyPointer is not implemented for Py and PyObject, we should fix this after #446.
It remove lots of Py::from_owned_ptr*s.

[konstin]

LGTM

[konstin]

Merging master should make travis ci pass again (or at least only fail for minimum nightly)

[kngwyu]

Now CI passed, thanks