Age | Commit message (Collapse) | Author |
|
|
|
|
|
Internally, strings were stored using Unicode code points. However, when
being written to disk with the `file:write_file()` function, there were
converted to ISO-8859-1. According to the documentation, that is because
the file module is bytewise-oriented: the conversion to another encoding
than ISO-8859-1 is the responsibility of the caller.
Using unicode:character_to_binary() permits the script to convert the
Unicode string to an UTF-8-encoded binary.
Without this patch, the added testcase would fail with the following
error:
gmake[3]: *** No rule to make target '(...)/erlang.mk/test/h��test_core_makedep_non_usascii_paths/deps/test_core_makedep_non_usascii_paths_dep/include/hello.hrl', needed by 'src/hello.erl'. Stop.
In this case, the path (passed from the Makefile to the `makedep.erl`
script) contains UTF-8-encoded `é` characters but it was converted when
doing the final file I/O.
|
|
They are already listed in `$^` and thus already taken care of. No need
to visit them twice.
|
|
To do that, we simply include the PID of the Make process in temporary
file.
We also compare the finished temporary file with the actual target file.
If they are different, the temporary file is moved to the final target,
otherwise, we just get rid of the temporary file. This way, the target
timestamp stays the same if the content does not change.
|
|
--first-parent is not available on CentOS 7.
|
|
|
|
... in addition to the `$(force_rebuild_dep)` function.
It's easier for the common use case than specify a function.
|
|
|
|
|
|
|
|
By default, Erlang.mk only builds dependencies once, the first time the
top-level project is compiled. A forced rebuild of all dependencies can
be forced by setting the `$(FULL)` variable to a non-empty value.
For developers working on the top-level project and one or more
dependencies at the same time, Erlang.mk now accepts a variable/function
called `$(force_rebuilding_dep)` which is called to determine if the
dependency passed as the first argument should be forcibly rebuilt.
It allows a developer to filter the dependencies he works on (so they
are always rebuilt) from other third-party dependencies he never touches
(i.e. they are only built once at the beginning).
The content of `$(force_rebuilding_dep)` is expected to be a shell
one-liner where the `$(1)` Make variable is set to the path to the
dependency.
Here is an example from the testsuite:
force_rebuilding_dep = test '$(1)' = '$(CURDIR)/deps/cowlib'
|
|
|
|
In the case of a multi-applications repository, if one application
depends on another one (using `$(LOCAL_DEPS)`), it will now be listed in
the `list-deps` targets.
Add `ssl` to the `$(LOCAL_DEPS)` in the testcase to ensure that
applications not in `$(APPS_DIR)` are not affected by this change.
Add a changelog entry.
|
|
Allow a lack of space in the front. This will of course need
to be handled better eventually.
|
|
... exactly as it is displayed for Git-based dependencies.
|
|
|
|
The issue was that when a new dep is added the last-makefile-change
update was done in the dependency and not in the top-level Makefile,
preventing the rebuild of the top-level project.
A test was also added to do the same when editing dependencies and
things seem to work properly in that case.
Thanks to robinchew on IRC for the great help debugging this.
|
|
For now, ~> is expanded to a full version only.
|
|
|
|
$ must be escaped as $$ in templates.
|
|
|
|
|
|
This solution should work better, previous one would sometimes
expand incorrectly and fail.
|
|
And use this in the tests to fetch Rebar only once and then
clone from the local repository. This should speed up tests
a little.
|
|
|
|
|
|
|
|
A release that gets removed in a ci:: preceding the running
of CI rules would end up not rebuilding it due to how a
warning was silenced when using ERLANG_OTP.
|
|
|
|
|
|
This allows refreshing tools at the same time Erlang.mk
gets updated.
|
|
V=2 only works against target commands, V=3 will also work
against $(shell ...) commands.
|
|
|
|
Unless it's a symbolic link, it's built directly, FULL=1 is set
or the file ebin/dep_built in the dependency is removed.
See the documentation changes for more details.
This provides immense build speed gains, for example on a
RabbitMQ project it went from 10s to 1s for the 2nd+ builds.
|
|
|
|
For example:
DEPS = elixir
ELIXIR_PATCH = 1
include erlang.mk
A language test suite will soon be added with a test for this.
|
|
The $(...) form was improperly converted. In addition it
must be changed into a $(shell ...) form.
|
|
|
|
Currently experimental. It'll enable lfe.mk when it encounters
the rebar3 plugin lfe-compile. Tested against some small library.
Note that it adds lfe as a build dependency with no version given
so the latest master will be fetched right now. This can be
overriden in the parent project.
|
|
Until it is found that one is needed to compile a project.
|
|
It's better than extending help:: because this gets printed
at the end instead of before Erlang.mk's own help.
|
|
|
|
When this option is used no include file is generated.
Therefore don't error out if no include file can be found.
|
|
|
|
|
|
When the $(MAKE) is inside an $(eval $(call ...)) then it needs
to be escaped as $$(MAKE).
When the $(MAKE) is inside a $(call ...) then I did not figure
out a way other than passing it as an argument.
When the $(MAKE) is inside many levels of $(call $(call ...))
it's easier to avoid it if at all possible, so I replaced the
rebar $(MAKE) call with ./bootstrap. I confirmed it works fine
on Windows as well.
|
|
Also remove the last instance of an "eval export" that pollutes
the environment and use the new core_render for it instead.
core_render does not include $(verbose) as this needs to be
different depending on who calls it. It's now set explicitly
everywhere required.
|
|
Fetching rebar is conditional depending on the project
being built, and if two projects require it and -j is
used this could mean the fetching happens at the same
time and one would fail.
|
|
I've reworked how the multi-apps projects are built. In particular
I've made sure the test build is made from the top-level once, and
then only tests are run on this build. It used to build multiple
times and some builds would not include test mode, not good. I've
also fixed issues with running tests in parallel.
All tests now pass with -j8 on my machine. It's possible more
issues remain that are not covered by tests yet though.
|