fix: timedelta parsing for int and floats

[tcleonard]

Version 11.2.0 introduced a regression when it comes to providing timedelta as floats or integers.

toto = env.timedelta("TOTO", default=30)   # does not works
toto = env.timedelta("TOTO", default=30.0)   # does not works
toto = env.timedelta("TOTO", default="30")   # works
toto = env.timedelta("TOTO", default="30.")   # does not works

This MR aims at making those 4 cases valid

[ddelange]

looks like a lot of logic was added just to bypass the parent class' validation (this line won't be hit anymore).

• a float as default value was previously slipping through the cracks (silent downcast to int before #366, a bug not a feature imo)
• a float-in-string as default value didn't slip through the cracks (expected behaviour to reject)

is this PR something that should rather fold into the bigger topic of #270 and/or #297 instead?

[ddelange]

opened a minimal version to avoid breaking change: #369

[tcleonard]

(this line won't be hit anymore).

It will if you provide something that is of other type than str, float, int, bool, timedelta.
So this is more a last option catch-all.

We could definitely do the more minimal change to get to the previously working (but unconsistent/unlogical behavior)... but I'm not sure what's wrong with this PR which supports the previously working pattern and actually fixes some of the inconsistent behaviors?

[tcleonard]

and actually, I think that #297 is not relevant anymore as it was already fixed (prior to this PR)

[sloria]

merged solution in #369

[tcleonard]

Sorry but I don't understand why this is closed @sloria . The other PR solves the breaking change that was introduced... but it persist the inconsistent behavior from before.
You can take the tests written in this PR and run them against the released version and you will see that it fails.

So why close this proposal of an alternative that makes the overall user experience better?

[sloria]

We can re-open this for now to resume discussion on the changes that go beyond the fix.

My reasons for closing--take them or leave them--were:

• Fix timedelta parsing breaking change #369 provided a more minimal change that addressed the immediate breakage. Because it had a minimal footprint, I could quickly review and release it.
• I'm very tired. I have limited time/energy to devote to OSS at this moment, as it's all volunteer work for me. I'm not using this project for an employer (because I'm unemployed!)

[ddelange]

I've opened an upstream fix marshmallow-code/marshmallow#2654

I think this can now be closed