BLD: travis seems to have not installed dateutil on py3.3

[hayd]

Just to note that this build failed, which is strange since the commit #4127 seems to only change one character in a docstring, I'm assuming it's a travis issue?

https://travis-ci.org/pydata/pandas/jobs/8765391

[jreback]

@y-p this is the first build after the change to use cached builds...take a look?

[jtratner]

I saw this too on a recent build

On Fri, Jul 5, 2013 at 9:33 AM, jreback notifications@github.com wrote:

@y-p https://github.com/y-p this is the first build after the change to
use cached builds...take a look?

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/4132#issuecomment-20519032
.

[cpcloud]

it looks like a pip issue

[cpcloud]

[cpcloud]

hm ...

https://travis-ci.org/cpcloud/pandas/jobs/8770106
https://travis-ci.org/cpcloud/pandas/jobs/8772752

[cpcloud]

think pip stuff needs to be installed "serially". sometimes when i do things like pip install dep1 dep2 dep3 only a few end up getting installed...kind of magical, never really looked into why but installing them one at a time always works

[jtratner]

I think it has to do with packages with compilation required (e.g., if you
try to install numpy with packages that depend on it, tends to fail). But
it's still voodoo like.

On Fri, Jul 5, 2013 at 4:32 PM, Phillip Cloud notifications@github.comwrote:

think pip stuff needs to be installed "serially". sometimes when i do
things like pip install dep1 dep2 dep3 only a few end up getting
installed...kind of magical, never really looked into why but installing
them one at a time always works

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/4132#issuecomment-20537677
.

The merge commit for the change is here : https://travis-ci.org/pydata/pandas/jobs/8749991
and it looks like dateutil installed fine. All the tests passed as well. The builds on my fork too.

This looks like a standard travis/pypi glitch to me and a coincidence. I'll take a look though,
maybe I did break something.

[cpcloud]

dateutil is fixed...more issues tho with np versions. see #4142

What was the underlying problem you fixed? the line you changed has been there for months,
and succeded on the original build. it also worked for all the other jobs except 3.3 in the failed
build hayd reported.

I can repro the problem with 3.3 now, though.

[cpcloud]

so u can repro it as of latest master, with the change i made?

i don't know what the underlying problem is. could be pip, could be travis...i've run into this issue before on my own machines where installing multiple packages at once sporadically worked. looping over the deps, though, has never failed.

fixing the underlying cause and not just putting a band-aid over the symptom is probably going to require delving into pip

i'm happy to revert this change if you want

I meant I could repro on hayd's commit. no need to revert if it works. I'm familiar with
the simultaneous deps problems you mention, though never intermittently. it consistently
fails in my experience.

pip (and pypa) is going through a lot of changes right now, if this is a bug in pip I'd like to
make sure it gets upstream.

[cpcloud]

hm. i might remembering incorrectly about the sporadic part. i might be able to look at pip this weeked