Asset pipeline broken for glyphicons on OpenShift (production deployment)

[brevilo]

When a basic Rails 4.1 application is deployed on OpenShift (production) glyphicon assets get precompiled with a digest suffix. When such an icon is used in a view, however, the related font assets are requested without that digest, causing the request to fail and hence the icons not to render properly.

As can be seen in the previous discussion leading up to this I'm pretty sure this is not caused by a flawed integration (like application.css.scss filename, missing @import or their order, //=require remnants).

Although this issue might not be caused by bootstrap-sass itself I'd like to understand what's required such that the correct asset URLs get used. Here I'm looking in particular at things like $bootstrap-sass-asset-helper and how these bits can be affected by a deployment on OpenShift.

Thanks

[dannyvanhoof]

Exactly the same issue on Heroku. Also here apparently not related to @imports or their order...

I managed to solve it by moving all my application dependent code from my bootstrap-and-overrides.css.scss directly into my application.css.scss file. Don't know though why this solves the issue

[brevilo]

Thanks Danny, but my example app above doesn't sport any overrides and uses only application.css.scss itself. But it's kinda "good" to see that I'm not the only one experiencing this issue. I'm really curious what's causing this...

[brevilo]

Ok, after I found that this issue also affects the font-awesome-rails gem I tried to tackle the problem from a different angle. Eventually it turned out that OpenShift doesn't explicitly set RAILS_ENV=production when precompiling the assets. While this omission still allows for generating the digest suffixes (hence my confusion about the effective environment), it prevents the app from requesting the assets with the correct URL (including the digest). Not sure why that is. I would expect either no fingerprinting/digests at all or the asset pipeline to fully work - not a mixture where one part behaves like being in the production environment but the other one doesn't! I can imagine that Heroku might do it the same way...

Anyhow, the easiest way to fix this problem on OpenShift is to provide the missing environment variable at all times (just to be sure). One can do that by logging into the gear and then creating a new file .env/user_vars/RAILS_ENV with the content production. I don't think that enabling this variable at all times should have any side-effects (does it?). It did instantly solve the problem for me.

There are obviously many others out there with the same problem, so I hope this helps!