The error bar does plot, but for some reason, it doesn't show the vertical connecting bar. You'll see if you do this:

fig = plt.figure()
ax = fig.add_subplot(111)
ax.errorbar([1.,2.,3.],[1.,1.,1.],yerr = [0.1,3.,0.4])
ax.set_yscale('log')
fig.savefig('test.png')

something like this outputs:
http://i.imgur.com/LOIi7.png

I think the problem lies in the fact that numpy computes log(1), in any base, equal to 0

Confirming. Pretty sure this is related to the fact that you are trying to plot an error bar down to log(0). This works if your not trying log(0). For instance:

ax.errorbar([10, 11, 12], [8, 9, 10], yerr=[0.5, 0.7, 0.9])

Looks fine. @M-funk: can you confirm what you are expecting to see? I can't see that there is an action on this issue (perhaps a warning might be suitable when you are at log(0))?

This can be worked around by doing

ax.set_yscale('log', nonposy="clip")

Maybe this should be the default for set_yscale? This seems to cause quite a few problems.

@neggert : Nice workaround.
Wish I had seen it before I wasted 1 hour to try and fix it ;)

From a beginners standpoint I would definitely vote for making it the default (unless there are obvious reasons not to do so).

👍 on making the default clip, with the appropriate API change warning.

Uhm... No, the errorbar should not be asymmetric when passing in log scale.
Please consider
http://faculty.washington.edu/stuve/uwess/log_error.pdf
If I understood correctly the documentation, I would consider nonposx="mask" as the best option of the default behavour.