Every directory with a jar.mn now has JAR_MANIFESTS defined in its
moz.build file.
We also removed the may_skip special consideration of jar.mn files
because this information is now available during tier traversal by the
reader courtesy of the variables being present in moz.build files.
JAR_MANIFESTS can now be defined in moz.build files. However, due to
limitations in rules.mk, only 1 file may be defined at a time. In the
future, this restriction will be lifted. But first, better support for
JAR manifests in the build config must be built.
rules.mk will be updated in the subsequent conversion patch so this
patch applied alone doesn't break the build.
There are two parts to this. The first is to add AndroidManifest.xml
as a dependency to the "no dependencies" ap_ built during packaging.
The aapt call requires it. So "no dependencies" is more accurately
"no *resource* dependencies".
The second is to avoid including the Android res/ directory in the
language repack step. What happens is that the l10n.py script sees
the Android res/ files left in the dist/ directory after unpacking and
expects to find them in the objdir. They're not there, so the script
fails. To avoid this, we delete them after unpacking. See the
comments in packager.mk describing this process.
Previously, the set of WebIDL example interfaces to build was hardcoded
in the mozwebidlcodegen Python package. Unfortunately, the example
interfaces in that set were dependent on test-only bindings files,
resulting in build failures when tests were disabled (because those
test-only bindings were defined in a test directory).
In this patch, we now declare example interfaces in moz.build files
so the set of example interfaces lives next to the set of defined
bindings files. The example interfaces are defined in a test moz.build
file - in the same file declaring the bindings files that contain the
interfaces - so the example interfaces only get picked up if the
corresponding bindings are defined.
We could probably switch WebIDL moz.build variables to use lists with
flags. That would be good followup fodder. For now, let's fix the build.
We now capture and print the number of Makefile.in and the number of
generated Makefile as part of config.status. This should give us a nice,
easy to extract metric going forward.
Pushing on a CLOSED TREE because this patch didn't deserve to get backed
out.
We now capture and print the number of Makefile.in and the number of
generated Makefile as part of config.status. This should give us a nice,
easy to extract metric going forward.
WebIDL build system integration has been rewritten from the ground up.
Changes:
* GlobalGen.py, BindingGen.py, and ExampleGen.py have been removed in
favor of mozwebidl.py.
* Static .webidl files are now processed directly in their original location
and aren't copied to the object directory.
* Generated events <stem>.cpp files are now compiled into the unified
sources. Previously, only the <stem>Binding.cpp files were compiled
into unified sources.
* Exported .h files are now generated directly into their final location.
Previously, they were generated into the local directory then
installed in their final location.
* The list of globalgen-generated files now lives in Python and isn't
duplicated in 3 places.
* The make dependencies are much simpler as a result of using a single
command to perform all code generation. The auto-generated .pp file from
code generation sets up all dependencies necessary to reinvoke code
generation and Python takes care of dependency management.
Build system developers commonly need to see what changes have on the
generated build files. We often put our objdir under version control and
diff commits before and after running config.status.
This patch adds a --diff option to config.status that will print diffs
of changes made during config.status. This functionality is implemented
on top of FileAvoidWrite, using Python's built-in diffing library.
While display of diffs is opt-in, diffs are always being captured when
config.status runs. There could be an unwanted performance regression
from this. Because diffs are only computed if files change and most
files don't change during most config.status runs, this greatly reduces
the surface area of the concern. The area for largest concern is clobber
builds. On my machine, I measured an increase of 0.2 to 0.3s from 2.0s.
While this is 10-15%, the total time is so small that I don't feel
snaking a "capture diff" flag through the build system is worth the
effort. This would make a decent followup bug if this turns out to be a
problem in the future.
I also snuck in a change to reindent all-tests.json because displaying
diffs for this massive 11MB all-in-one-line JSON file results in an
extremely large string being printed to my terminal.
This patch adds pattern matching entries to install manifests. We store
metadata necessary to construct a pattern match at a later point in
time. When we convert the install manifest to a file registry, we
resolve the patterns using FileFinder.
The build config logic has been updated to store support-files values as
pattern entries. This should resolve the clobber needed issue and make
the local development experience more pleasant as well.
This patch adds pattern matching entries to install manifests. We store
metadata necessary to construct a pattern match at a later point in
time. When we convert the install manifest to a file registry, we
resolve the patterns using FileFinder.
The build config logic has been updated to store support-files values as
pattern entries. This should resolve the clobber needed issue and make
the local development experience more pleasant as well.
