Closed garborg closed 8 years ago
It looks like some of these Travis failures are going to persust until nolta/Winston.jl#236 gets into a tagged release.
If there are remaining failures then, following Winston's (.travis.yml) lead on sudo: require
and/or xvfb-run
may be required.
I'm sorry I'm just seeing this, Sean. I missed the notification.
I like the idea of using Inf
, that seems to make sense. 0.0
could be misleading.
Can you walk me through a couple of the travis changes you made?
First, is "julianightlies" no longer a thing? Or does Travis now support Julia as a language and so the repositories are already available?
And did you remove all the installs because they should be coming through now as Winston's dependencies? (This makes a lot of sense, there must have been a reason I was doing this originally.)
I'm worried this is going to break on Travis' new container-based build system, which doesn't give sudo access. This script won't tell Travis that sudo is needed to install some of the dependencies, so it won't run on the legacy build system. (scratch that -- looks like it's still on the legacy system).
Ah, I see. "Julia for Travis-CI is not officially supported, but is community maintained." Cool! Ok, I understand all the changes now. Disregard my previous comment.
Hi Ted,
I've added an explicit sudo requirement in case the default environment changes. And the previous failures from calling xvfb
on the new osx builds have been resolved. Now there's just the segfault on 0.4, which I want to say "seems unrelated" but those words get eaten too often, so when I have time, I'll see if the current repo state triggers the same segfault.
Nightlies are still a thing, but 0.5 is changing so quickly that I don't imagine most packages supporting it for a couple months or so, and with our dependencies it would be tough to keep pushing support upstream for now.
I moved the incomplete travis config update commit to #77, so this PR should be an uncontroversial improvement now.
Thanks, Sean! I've been out of town the last few days, sorry about going silent.
No problem, Ted. I was on the road myself for the last few days and didn't notice.
distance
used to return0.0
for empty routes (what the route funcitons return when they can't find any route), but I accidentally made it an error during the switch to useGeodesy.jl
.I've got it set to
Inf
, because it matches what the route functions return for distance/time, but I'd be happy to switch it back to0.0
if that's preferable.