Age | Commit message (Collapse) | Author |
|
|
|
we also use LOCAL_DEPS at the top-level, if it exists, to determine
which local apps to compile, in the interest of saving compile time
for projects that have many apps and multiple release configurations
|
|
`rm deps.log` was not being executed, because of 1) the ifeq around
it, checking for an empty ALL_DEPS_DIRS, which i believe was a logical
error; and 2) deps depended on apps, which means apps were compiled
before its recipe was executed, and the `rm deps.log` would be
executed after it had been written to by recursive make of the
apps.
the important reason to remove deps.log is so that dependencies get
remade when they change. otherwise you'll get mysterious errors about
missing dependencies if they need to be rebuilt, or worse, have wrong
stale beam files bundled in your release.
tests core-deps-apps and core-deps-apps-only now manually clean the
cowlib dep and rebuild at the top-level to make sure cowlib gets
rebuilt. both tests indeed fail without this fix.
this attempts to fix #1 by removing the ifeq, and #2 by having deps
and apps depend on deps.log, with deps.log 'built' (but actually
removed) by a double-colon rule with no prerequisites (which means its
recipe always be run).
|
|
This allows all plugins to append to existing variables
and makes it easier to augment the functionality of the
core plugins.
|
|
Because we were unconditionally including the makedep file,
it would be built even if the target was unrelated (like
"make help"), and it would also trigger building of xrl
and other files that produce Erlang code.
Instead the makedep file is now included as a wildcard:
if it exists it gets included, if it doesn't it's not,
and if it's created in the middle of execution it gets
included only then.
More details on the technique can be found at
http://make.mad-scientist.net/papers/advanced-auto-dependency-generation/
A test has been added to make sure the behavior does not
come back at a later time.
|
|
`ERLC_ASN1_OPTS` can be used to pass compiler options when compiling
ASN.1 files. By default, Erlang.mk will leave this empty, but it can
be redefined in a Makefile.
|
|
|
|
Erlang.mk projects either have a .app.src, or just the Makefile
with an optional .app file for compatibility.
|
|
If we have a .app and a .app.src we delete the .app.
If we have a Rebar project with a .app file, rename it to
.app.src and continue with autopatch.
If we have a non-Rebar project, leave the .app file, and then...
For all cases, delete the ebin/ folder after autopatch completes.
|
|
|
|
|
|
This will make sure we don't look twice.
Also make a single call to find the common file formats
rather than many calls.
|
|
The initial documentation is a bit rough. It's getting late and
I'd like to have this in sooner rather than later. :-)
|
|
The previous approach spawned a shell for every single source files. As
we disable parallel make jobs, this is very time consuming.
The new approach uses an intermediate timestamp file in $(ERLANG_MK_TMP)
to record the last modification of any of $(MAKEFILE_LIST) and touch
all source files in one command. Then, the .app file depends on this
timestamp file.
We test the existence of this timestamp file: if if doesn't exist, don't
touch source files, they will be built anyway.
$(PROJECT).d now depends directly on $(MAKEFILE_LIST); before, this
dependency was indirect through $(ERL_FILES). Also, once $(ERL_FILES)
were touched, we do the same with $(PROJECT).d because there is no need
to regen it because of this.
|
|
Error has been detected during the nightly packages build.
|
|
|
|
After a lot of soul searching and experiments, I am opting
to using curl for fetching files. It works with very little
manipulation on all supported systems. Some might need it
installed, but most come with it.
While wget was working, it had TLS validation deactivated
because some users didn't have CA certificates on their
system or wget was not configured properly. This does not
seem to be a problem with curl as far as I tested, in
particular on OSX.
The fallback to use Erlang was also insecure, and probably
not the best idea. Let's keep things simple. Curl is required.
|
|
.. to fetch and list deps recurvively.
Therefore, they work on direct dependencies, dependencies' dependencies,
and so on. Nothing is built with those targets.
The following targets are also available to work on other kinds of
dependencies:
make fetch-doc-deps
make fetch-rel-deps
make fetch-test-deps
make fetch-shell-deps
make list-doc-deps
make list-rel-deps
make list-test-deps
make list-shell-deps
In all cases, they include "normal" and eg. "test" dependencies of the
top-level project, then only "normal" dependencies' dependencies.
It's possible to include several kinds in one go:
make fetch-deps DEP_TYPES='doc test'
make list-deps DEP_TYPES='doc test'
As it may be difficult to use the output of `make list-*` because
the list could appear after other targets output (like dependencies
fetching), the list is available in files pointed by the following
variables:
$(ERLANG_MK_RECURSIVE_DEPS_LIST)
$(ERLANG_MK_RECURSIVE_DOC_DEPS_LIST)
$(ERLANG_MK_RECURSIVE_REL_DEPS_LIST)
$(ERLANG_MK_RECURSIVE_TEST_DEPS_LIST)
$(ERLANG_MK_RECURSIVE_SHELL_DEPS_LIST)
Internally, `list-deps` is simply implemented on top of `fetch-deps`:
the latter fills a sorted log with all the directories it traversed.
`list-deps` finally just prints the log file to stdout.
Fixes #560.
|
|
This allows to add standard keys which do not have a corresponding
`PROJECT_` Makefile variable, as well as non-standard keys.
|
|
|
|
|
|
A warning will be displayed for the time being while we keep
compatibility.
Also improves installation docs for Unix.
|
|
|
|
Fix #587
|
|
|
|
|
|
This change makes it possible to override the ERL_CFLAGS and ERL_LDFLAGS
Makefile variables in project dependencies. For the main project, the
ability to override variables associated with C compilation already
existed in plugins/c_src.mk. With this change, the C compiler flag
semantics are similar for both the main project and dependencies.
This is important for cases where the automatically determined values
for the ERTS include and library directories are incorrect. This happens
in cross-compiled environments where they point to the host's include
and library directories. However, the desired behaviour is to have them
point to those that have been cross-compiled for the target.
|
|
|
|
This reverts commit d5c32474e0690a32371fb8fb0b5d2db2c567269b.
|
|
This should fix some Windows issues.
|
|
Replicating the behavior for deps/*.
With refactoring from Loïc.
|
|
|
|
|
|
Create ebin dir for all apps_dir apps before building any of them.
|
|
|
|
|
|
|
|
Extended Mode uses { and } as special characters. Making grep
implementations other then GNU grep fail (see
http://www.gnu.org/software/grep/manual/grep.html#Basic-vs-Extended).
|
|
|
|
|
|
Fixes build of a few projects on Windows.
|
|
|
|
For easier reusability in Erlang.mk and plugins.
|
|
When preparing erlc_opts, remove all comma/whitespace pairs instead of
just the first
|
|
Fixes syslog.
|
|
Also includes a fix for multi-application repositories.
|
|
Rebuilding cleanly first prevented rebuilding in test-mode again.
|
|
This makes i18n work. It worked before, but I didn't know,
and broke it. Next commit will add i18n to the index to
ensure this doesn't happen again.
|
|
One case was added, where a -compile attribute specified a list
that included a parse_transform. This affected alog, for example.
The COMPILE_FIRST variable is now filled from a digraph. This
allows us to keep track of the dependency tree properly, rather
than rely on luck. This affected erlando.
With both of these changes in, a large chunk of the rebar
autopatch code can be removed, the part that concerned the
auto detection of file dependencies.
On the other hand, we still need to honor the erl_first_files
configuration value, otherwise the parse_trans project fails
to build. While it would be possible to detect these dependencies
automatically too, it's probably too complex to bother, at least
for now.
|
|
This allows us to avoid autopatching when the Makefile of an
Erlang.mk project contains the word "rebar".
Had to update elvis and xref_runner to master for now because
of incompatibilities. The problem went unnoticed for a while
because they were autopatched.
|