Log axvline

[pelson]

Fixes the exception in from #1246 of:

ValueError: It is not possible to compute transA - transB since transB 
cannot be inverted and there is no shortcut possible.

The fix was simple enough (added the has_inverse property to the Log transforms). The generated png for the tests looks perfect, unfortunately the PDF and SVGs are not. Strangely, when I run this from a script rather than from the tests, the PDF and SVGs are perfect, but the png produces a line in the wrong place. If anyone has any ideas what may be going on there, it would be greatly appreciated.

As it is, this is a bona fide bug fix, but there still exists some weirdness to investigate.

[dmcdougall]

@pelson Nice!

[dmcdougall]

@pelson Which backend are you using for the png?

[pelson]

Now there is a good question. In the script, I have 'TkAgg'.

The pngs are bad when created from the Agg backend. Hmmm, curious. What is also funny is that the line produced in the test for SVG and PDF is different to that in the script. I'm sure I'm doing something daft.

[pelson]

Needs removing. Its not this that's causing the problem though.

[dmcdougall]

Now there is a good question. In the script, I have 'TkAgg'.

The pngs are bad when created from the Agg backend.

How? The line is in a different place? Or is it bad quality?

Can you compare the line created with the agg backend and the line created with the cairo backend?

What is also funny is that the line produced in the test for SVG and PDF is different to that in the
script. I'm sure I'm doing something daft.

In some sense, that kind of makes sense. Doesn't the testing framework nuke all your rc defaults?

[pelson]

The line is in a different place?

Completely.

[dmcdougall]

The line is in a different place?

Completely.

Bugger.

[pelson]

:-) indeed.

Ok. So I'm going to take stock, and have a good think about this one. The order of saving the results is important, so I think there is some caching issues going on. I'm happy to spend some time on this in the next couple of days to track this down.

[pelson]

This issue should now be completely resolved by this PR. The only remaining concern is getting the text in the image to pass on travis. I will add a new commit which removes the text from the images.