ENH: using python's warnings instead of prints in fitpack.

[Dapid]

When fitpack detects a recoverable problem, it will signal it with
a RuntimeWarning instead of printing to stdout. This way the output
can be programatically silenced or made exceptions.

There is one failure on Travis, but I don't see what has failed.

[WarrenWeckesser]

The failure of job 3059.3 is the 50 minute timeout that we've been seeing regularly. The failure of 3059.1 is something new. It looks like Travis stops the build when the number of output lines hits 10000. I got the same failure with #4004

[insertinterestingnamehere]

The Travis build only shows the first 10000 lines of output in their main view. If you view the log as a text file, it will show the remaining lines. The link for that is on the right side a little down from the top. It has four horizontal bars. The error shown in both of those build failures is

scipy/ndimage/src/ni_filters.c:386:5: error: ISO C90 forbids mixed declarations and code [-Werror=declaration-after-statement]

[WarrenWeckesser]

@insertinterestingnamehere Thanks! I was just about to update my comment to note that the log might be truncated. Good to know that is actually what is happening.

[insertinterestingnamehere]

No problem. Looking back, I think merging #3527 broke this. That's just a guess though.

[WarrenWeckesser]

Yup--you're one step ahead of me again. :)

[Dapid]

Should I add some tests to ensure the warnings are raised? Ot is it too runtime dependent?

Now I think about it, the docstrings should be modified too. What about this:

    quiet : bool
        Non-zero to suppress messages.

becoming this?

    quiet : bool
        Non-zero to suppress messages. Else, they will be issued as `RuntimeWarnings`.

[ev-br]

You can add a test with mest=0, so that the intention is clear. Re: docstring, I'd actually add that this parameter is deprecated and we suggest its not used in new code (use standard python warnings filters instead)

[ev-br]

Nitpick: IMO it's better to keep importing print_function even if neither print() function nor print statement should be used in the library code.

[Dapid]

Why is that?

[argriffing]

It is boilerplate at the top of every python file for python3 compatibility. I think there's even a bot that enforces it, or that added them at some point.

[ev-br]

LGTM modulo nitpicks. Adding a test for mest=0 is not essential I think.

Restarted the tests, TravisCI is green.

[coveralls]

Coverage increased (+0.02%) when pulling ad056f9 on Dapid:warnings into fdddfc3 on scipy:master.

[Dapid]

I have edited the docstrings and put back the import. I am not sure about how to do the mest=0 test.

[coveralls]

Coverage increased (+0.02%) when pulling ae276ab on Dapid:warnings into fdddfc3 on scipy:master.

[ev-br]

There's always something as simple as assert_warns and

>>> x = np.arange(8)
>>> y = 2.*x - 3
>>> tck = splrep(x, y)
>>> sproot(tck, mest=0)
Warning: the number of zeros exceeds mest
array([], dtype=float64)

But IMO this is borderline useful at best.

TravisCI is green, merging. Thanks @Dapid, Warren, Ian, Alex.