PGF backend, fix #1116, #1118 and #1128#1124
Conversation
|
On a machine that still has the "too old" TeX, I now get this, which is correct -- they are known failures (and I've turned off the KnownFailure class here just so I can see exactly where they are coming from). However, on a machine where I upgraded the TeX to TeXLive 2012, I'm getting this: I've put my PDF files up here: http://mdboom.github.com/scraps/pgf_pdflatex.pdf http://mdboom.github.com/scraps/pgf_rcparams2.pdf As for the Python 3 stuff, I plan to tackle #983 today, and that should hopefully get the python 3 tests running for you. |
|
Downloading pgf_rcparams2.pdf doesn't work for me, but from the pgf_pdflatex.pdf you uploaded I get that the default fonts of our latex environments are slightly different. The difference is very minimal and shouldn't cause the test to fail. When I wrote the tests I lowered the default tolerance of 1e-3 to 1e-4 because the test didn't recognize when I changed the fonts from serif to sans. Seems that tol=1e-4 is too restrictive. I reverted the tolerance back to 1e-3. Does this work for you? |
|
Oh.. and that knownfailureif-decorator confuses me.. for some reason, the pdflatex test actually worked with your old tex distribution.. and the test raised a KnownFailureDidNotFailTest exception. What I'm actually looking for is a decorator that doesn't try to verify if the test really fails and blames me if it succeeds :). |
|
If I change the tolerance to 5e-3, it works for me. |
|
So tol=1e-3 didn't work for you neither? I'm using 5e-3 now. Furthermore, I discovered SkipTest in nose.plugins.skip which does exactly what I wanted. It doesn't require that a test that is marked as known-to-fail actually fails. |
|
Ahh ok.. I now see whats the big difference between those images.. Its just ghostscript being horrible at rendering text for some (many) fonts. I checked the PDFs using the gnome viewer and gimp (both cairo based) and saw no difference. The gs converted PNG files both look horrible but in a different way. |
|
I first tried a tol of 1e-3, but that was still too strict. 5e-3 was the lowest I could go to get the tests to pass. That's a bummer about gs -- we could probably move to another renderer down the road, but not for this release. I think we just raise the tolerance for now. |
|
I finished running the tests using python3. Some additional ResourceWarning messages are emitted when using py3. I got rid of the ones being traced back to backend_pgf.py. These remain: Nothing to be concerned about though. I think these fixes are good to go. |
|
Thanks for working through this so quickly! |
Commit 86ac6ce undid the changes from commit d527799 to change the behavior of shift-enter in Jupyter notebook, as discussed in matplotlib#1124, matplotlib#4758, and matplotlib#6752. With the changes here, shift-enter now selects the next cell, executes it, and then selects the following cell.
This PR contains the fixes for #1116, #1118 and #1128.