Make the format strings compatible with Python 2.6

[vgrigoriu]

This should restore compatibility with Python 2.6, current version fails with:

Traceback (most recent call last):
...
  File "/usr/lib/python2.6/site-packages/prometheus_client-0.5.0-py2.6.egg/prometheus_client/exposition.py", line 194, in write_to_textfile
    f.write(generate_latest(registry))
  File "/usr/lib/python2.6/site-packages/prometheus_client-0.5.0-py2.6.egg/prometheus_client/exposition.py", line 119, in generate_latest
    output.append(sample_line(s))
  File "/usr/lib/python2.6/site-packages/prometheus_client-0.5.0-py2.6.egg/prometheus_client/exposition.py", line 86, in sample_line
    s.name, labelstr, floatToGoString(s.value), timestamp)
  File "/usr/lib/python2.6/site-packages/prometheus_client-0.5.0-py2.6.egg/prometheus_client/utils.py", line 21, in floatToGoString
    mantissa = '{}.{}{}'.format(s[0], s[1:dot], s[dot+1:]).rstrip('0.')
ValueError: zero length field name in format

[brian-brazil]

Can you add the DCO?

[vgrigoriu]

Done.

[brian-brazil]

The tests are failing on 2.6, are you sure this is broken?

[vgrigoriu]

Looking into the failing tests.

[vgrigoriu]

So, I think I figured things out. The tests fail in Python 2.6 because of this line.

The test tests that all those values generate ValueErrors. This passes in Python 2.7 because floatToGoString('1.5555555555555201e+06') is '1.55555555555552e+06' -- see this line.

In Python 2.6, without my changes, the test passes because this line, coincidentally, throws ValueError (because 2.6 requires numbers for placeholders).

With my changes, this line no longer throws ValueError and it happens that floatToGoString(1.5555555555555201e+06) is '1.5555555555555201e+06' (the same), so the expected ValueError is not thrown.

The difference in the result of floatToGoString is that in 2.6 repr(1.5555555555555201e+06) is '1555555.5555555201' and in 2.7 it is '1555555.55555552'.

Obviously I don't know why this specific value was chosen for this test, but maybe we should remove it? What do you think @brian-brazil?

[brian-brazil]

Odd that CI let this though.

Obviously I don't know why this specific value was chosen for this test, but maybe we should remove it?

It's there specifically to catch this. This sounds like https://bugs.python.org/issue7117 or https://bugs.python.org/issue7632

[vgrigoriu]

Odd that CI let this though.

This passed in CI for 2.6 because the only test path exercising this code happens to be in a test that expects ValueError.

I pushed a change that makes floatToGoString work the same for the existing test cases in 2.6 and 2.7.

[brian-brazil]

I'm not sure this is correct, as it's the underlying algorithm that's not right.

[vgrigoriu]

Are you saying that there's no easy way to preserve compatibility with Python 2.6?

[brian-brazil]

Possibly not, we'd need our own floating point renderer for it.

May I ask why you're still on 2.6?

[vgrigoriu]

For my sins :-P

I'll close the PR then, I'm good with using 0.4.2 until we can upgrade to Python 2.7.

[brian-brazil]

We should still fix the format bug on 2.6

[vgrigoriu]

So what do you want to do about the test failing on 2.6?

[bz2]

Can skip the test on 2.6 only with a message referencing the floating point formatting compatibility issue.

[brian-brazil]

Thanks!