Newer version of docker no longer have the /.dockerinit file present
when the container is executing, so that is no longer an option.
When executing in a container, we do know that we will be executing
as pid 1 -- this is probably not ever the case on the host system.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
This moves things around a fair bit in the .travis.yml to be
more logically ordered and adds support for publishing docs
(on master) to gh-pages. Code coverage is also calculated
and published to coveralls.io.
Travis is able to write to gh-pages on the github repository
via the GH_TOKEN environment variable stored in the travis
settings (marked as private). This token is generated from
the github user 'rust-nix-bot' in order to avoid given access
to other repos associated with an actual user's account (this
practice is recommended by Github).
Other changes worth noting:
- We now test fewer intermediate versions of rust
- The existing logic to publish to Carl's S3 bucket have been removed.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
Previously, std for x86_64 and i686 was included in the base
image. To reduce size, those contents are now in the x86
image instead.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
We are using containers for a majority of builds again, so perform
those builds on more platforms (as travis seems to handle this much
better).
For OSX, we now testing i686/x86_64 on 1.1.0 as well for added
coverage.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
Somewhere in the last set of changes, a few references to
`test_nix` were not removed. Since that function doesn't exist,
the tests would never be run on those targets.
This script is only used when testing locally right now.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
On the host, some undesired things could end up happening if this
is executed, so we now exit and log if that is attempted. In addition,
we now backup the previous cargo config if present. This will never
be the case in the current build setup, but that could change if
we decide to do builds in a data volume in the future.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
For platforms where the host ${CC} is appropriate, use the default
implicitly rather than matching it explictly. This will allow for the
llvm linker to be used if/when that happens.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
All of the platforms tested currently fail in some way currently, but
the infrastructure itself appears to be working. As we have moved to
the legacy infrastructure (required in order to use docker), the tests
do run slower now. To compensate for that, i686 builds are only done on
stable and we do not test every version between the current stable and
the oldest version we support.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
The rust-cross image is no longer all-in-one but broken out by architecture
class (or platform in the case of android).
Signed-off-by: Paul Osborne <osbpau@gmail.com>
We already need to test cargo about gcc locations so it can perfom
linking -- we just parse this information out to tell rust where
gcc is for complation (nix-test compiles some C code for testing).
Signed-off-by: Paul Osborne <osbpau@gmail.com>
Later on, it will probably make sense to stop on failure -- given
that most targets currently fail, however, this works a bit better
for the time being.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
There is now a published docker image (if a bit large) making
it possible for others to start using the new test ingfrastructure.
There seems to be no shortage of compile problems right now.
Signed-off-by: Paul Osborne <osbpau@gmail.com>
These scripts represent a rough first cust at adding support for
locally building nix and its tests for a large number of rust
versions and architectures. Currently, this mostly helps with
testing that nix compiles for various architectures (to the point that
there are a mound of problems that need to be addressed). Once these
are sorted, the code for actually running the tests for various
architectures will be approached (this will require qemu and will
still not target all architectures).
Signed-off-by: Paul Osborne <osbpau@gmail.com>
