All Meteor packages implicitly depend on the meteor package, and the
meteor-base package implies the meteor package for most Meteor apps, but
the new minimal skeleton does not use meteor-base, so minimal Meteor apps
were not directly depending on the meteor package.
The only reason this mattered was that the meteor package registers a
default compiler plugin for CSS files, and compiler plugins only apply if
an app or package directly depends on them (or depends on a package that
implies them, such as meteor-base).
In other words, this change reenables support for raw CSS files for
minimal apps.
The only new functionality is the res.setHeader("Vary", "User-Agent")
call; the rest of the changes in this commit are just reformatting.
@CaptainN I think this should be enough information for well-behaved
browsers and CDNs to do the right thing, though I agree it's worth
documenting.
This still leaves web.cordova as the only architecture whose URLs get
prefixed (with /__cordova/), but the implementation better reflects the
special status of web.browser and web.browser.legacy as architectures that
the webapp and dynamic-import packages understand how to disambiguate
using the isModern test from the modern-browsers package.
Now that webapp can differentiate between modern and legacy browsers when
serving static files, without relying on URL prefixes, in principle we
shouldn't have to use prefixes for any URLs except Cordova ones.
Since legacy assets are no longer disambiguated from modern assets with a
/__browser.legacy/ URL prefix (#9776), we need a principled way to handle
collisions in the staticFiles registry.
In many production systems logs from the application will be forwarded to various logging systems, sometimes third party ones. All credentials in your settings will be leaked if you have any errors in your json.
Feature request can be found here
https://github.com/meteor/meteor-feature-requests/issues/293.
* add resource-file to mobile-config
* Fix typo in addResourceFile documentation header
* get filename with path.parse instead of split
* Code formatting adjustments
This reverts commit 3062995801.
The tests exhibited on AppVeyor showed some very clear failures in the way
that we're killing PIDs on Windows using `taskkill`. I haven't actually
checked, but I find it virtually impossible that the PID of the process to
kill (the 5th test run) happened to be 3000.
Our current testing on Mongo is less complete than it could be and these
tests, particularly the `mongo with multiple --port numbers` test which does
actual writes to the database, should help with that.
Note that these tests are separate from the `mongo` package tests which are
much more extensive and test more advanced mongo functionality, but it's
good to have some basic Windows tests in place within the `self-test`
infrastructure.
cc @benjamn
