Revert "Use atleast_3d instead of ensure_3d" #5241
Conversation
This reverts commit 2f5a633. numpy.atleast_3d does not convert an empty array properly.
|
proposed fix for issue #5185 |
|
Cc: @jkseppan, as the author of the original commit. |
|
The doc failure is fixed on 1.5.x, will merge over now. I am still unclear on why this results in the intermittent failure and I could not reproduce this in linux. |
|
I'm fine with this going in, but it sounds odd that this would fix an intermittent failure. Is the use of |
|
@jkseppan Reading uninitialized data sounds like the only explanation I can think of. Unfortunately I don't have any experience with running valgrind or such on python extensions |
|
EDIT: Sorry, missed your explanation in #5185. Still puzzling, though. |
|
I believe it is due to out-of-bound memory access which is causing this random behavior. I tried to give some more insight in #5185 on what is happening in the path c++ extension code where it goes bad. |
|
Superceded by #5274. @cimarronm: Thanks for all your help identifying the issue! |
mathtext creates Python dictionaries for the charmap and inverse charmap for each font. This turns out to be unnecessary: 1) freetype has an API to do a charmap lookup that is faster than a Python dictionary 2) The inverse charmap isn't really necessary if we convert the latex_to_bakoma to use unicode character points rather than glyph indices. This should have a large impact when matplotlib#5241 is merged with larger fonts.
mathtext creates Python dictionaries for the charmap and inverse charmap for each font. This turns out to be unnecessary: 1) freetype has an API to do a charmap lookup that is faster than a Python dictionary 2) The inverse charmap isn't really necessary if we convert the latex_to_bakoma to use unicode character points rather than glyph indices. This should have a large impact when matplotlib#5241 is merged with larger fonts.
mathtext creates Python dictionaries for the charmap and inverse charmap for each font. This turns out to be unnecessary: 1) freetype has an API to do a charmap lookup that is faster than a Python dictionary 2) The inverse charmap isn't really necessary if we convert the latex_to_bakoma to use unicode character points rather than glyph indices. This should have a large impact when matplotlib#5241 is merged with larger fonts.
mathtext creates Python dictionaries for the charmap and inverse charmap for each font. This turns out to be unnecessary: 1) freetype has an API to do a charmap lookup that is faster than a Python dictionary 2) The inverse charmap isn't really necessary if we convert the latex_to_bakoma to use unicode character points rather than glyph indices. This should have a large impact when matplotlib#5241 is merged with larger fonts.
mathtext creates Python dictionaries for the charmap and inverse charmap for each font. This turns out to be unnecessary: 1) freetype has an API to do a charmap lookup that is faster than a Python dictionary 2) The inverse charmap isn't really necessary if we convert the latex_to_bakoma to use unicode character points rather than glyph indices. This should have a large impact when matplotlib#5241 is merged with larger fonts.
This reverts commit 2f5a633.
numpy.atleast_3d does not convert an empty array properly.