| Age | Commit message (Collapse) | Author |
|---|
|
The tests were waiting for the test group to finish before
they could continue with the next test group. Now "core"
and "all" targets directly depend on individual test cases,
allowing parallel Make to get to the next tests quicker and
removing 1/3rd of the total run time.
make check -j8 -k 5790,16s user 1207,08s system 627% cpu 18:35,49 total
make check -j8 -k 6250,13s user 1326,77s system 972% cpu 12:59,16 total
|
|
And use this in the tests to fetch Rebar only once and then
clone from the local repository. This should speed up tests
a little.
|
|
This allows refreshing tools at the same time Erlang.mk
gets updated.
|
|
|
|
|
|
|
|
Not only simpler but also faster, especially on Windows.
|
|
This way if for some unknown reason $(APP_TO_CLEAN) isn't
defined, we don't delete the entire world.
|
|
If the local build.config has changes compared to upstream's
master branch, and $(ERLANG_MK_COMMIT) also brings changes to this
file, copying the local file to the checkout before switching to
$(ERLANG_MK_COMMIT) caused git-checkout(1) to abort with:
error: Your local changes to the following files would be overwritten by checkout:
build.config
Please, commit your changes or stash them before you can switch branches.
Aborting
While here, fix `core-upgrade-custom-repo` test case: without switching
back the alternate erlang.mk repository to its master branch, the clone
always got the test-copyright branch by default, making the
ERLANG_MK_COMMIT check ineffective.
|
|
|
|
This fixes an issue on OSX where the OSX make 3.81 has a bug with
multilines single quoted arguments to commands: it strips the
backslash and newlines entirely. This makes it impossible to use
sed -i properly. Since this is the test suite, we can instead use
perl -ni, which can do the same and is available and work fine on
all tested platforms.
|
|
Use one file per topic, except for "uncategorizable" tests.
|
