We build gcc after clang, and extract libgcc libraries and libstdc++
headers from gcc and place them in the clang installation directory in a
way that clang favors before it searches the system for libraries and
includes.
This patch migrates moz_non_memmovable into the CustomTypeAnnotation
framework; bonus side-effects are more helpful error messages and less
code duplication.
Its only purpose is to disable PGO. Where that was not already explicitly done,
or irrelevant (because the directory only contains python), I disabled it in
moz.build.
This included adding the slugid 1.0.6 python source code in /python since slugid
is now a dependency of the ./mach taskcluster-graph command, as well us updating
references that used it. Previously the implementation was in-tree.
The patch removes 455 occurrences of FAIL_ON_WARNINGS from moz.build files, and
adds 78 instances of ALLOW_COMPILER_WARNINGS. About half of those 78 are in
code we control and which should be removable with a little effort.
This paves the way for defining additional Android packages in
moz.build, which is a step toward moving the special
mobile/android/base/Makefile.in aapt invocations into the generic
java-build.mk framework.
The new variables are both passthru variables for now: in the future,
we'll roll them into some aggregate Android APK definition.
It's worth noting that references to the variables in Makefile.in
files are only defined after including rules.mk (and thereby
backend.mk). This only required a few changes in the tree but it
confused me for some time.
GeneratedJNIWrappers.h was updated in bug 1192079 to use inherited
constructors, which is a gcc 4.8 feature. Many people are still using an
older version of NDK which only comes with gcc 4.7.
To reduce verbosity of the generated code, this patch makes the code
generator use unqualified names when possible, e.g. use State::Ref
instead of GeckoThread::State::Ref. To accomplish that, function
prototypes now use the C++11 -> syntax for return types.
Currently, when we generate JNI wrapper for an inner class, the
resulting C++ class will not actually be a nested class of the enclosing
class. As a result, the class can be confusing to use. For example,
wrapping Java class GeckoThread.State results in two unrelated C++
classes, GeckoThread and State, and it'd be confusing to use State by
itself.
This patch adds support for inner classes. We start by scanning only for
top-level classes, and when processing each top-level class, we
recursively scan for inner classes through
JarClassIterator.getInnerClasses() and CodeGenerator.generateClasses().
For each Java inner classes, the resulting C++ class will be a nested
class. For example, wrapping GeckoThread.State will produce
widget::GeckoThread and widget::GeckoThread::State.
We have had singular ANDROID_ASSETS_DIR in Makefile.in for a while.
Fennec itself does not use the existing Makefile.in Android code, for
complicated historical reasons.
This makes the existing variable moz.build-only; generalizes the
existing variable to an ordered list; and adds the equivalent use of
the new list to the Fennec build, with a simple example asset.
This patch also updates the packager to include assets packed into the
gecko.ap_. Without the packager change, the assets/ directory in the
ap_ gets left out of the final apk. This whole approach is totally
non-standard but is more or less required to support our single-locale
repack scheme.
The PR was fixed in early 2011. clang 3.3, the oldest version of clang
that we build with, was released in mid-2013. It's safe to say that all
versions of clang now have this fix, and we can delete the check.
The way UPLOAD_EXTRA_FILES is currently exported makes the files added
to the list added in each directory that is being traversed recursively
because of the += in .mozconfig.mk, combined with "export".
The easy way out is to remove the export altogether, but being unsure of
the side effects of putting all non exported mk_add_options variables
in .mozconfig.mk, limit this to UPLOAD_EXTRA_FILES.
When switching between Gtk+3 and Gtk+2, config.cache will contain a PKG_CONFIG
that may not be suitable for the build:
- after a Gtk+2 build, config.cache will point to the system pkg-config, which
doesn't like the pkg-config files in the Gtk+3 tooltool package.
- after a Gtk+3 build, config.cache will point to the Gtk+3 tooltool package's
pkg-config, which is likely not there in a Gtk+2 build.
Setting PKG_CONFIG avoids all config.cache considerations altogether, so set it
appropriately for both cases.
LeanSanitizer reports two kinds of leaks: direct and indirect. A
leaked block that is pointed to by another leaked block is an
"indirect leak", while one that isn't is a "direct leak". Often,
indirect leaks are just things entrained by the "real" leak, but if
two leaked blocks are in a cycle, then they both end up being
indirect, so we need to report them, too.
This patch makes it so that indirect LSan leaks are treated the same
as direct leaks by Mochitests, which means they will turn the tree
orange. There are a few existing indirect leaks of various severity,
so I had add some suppressions. See those bugs for more details.
The API version detection functionality was broken in SDKProcessor
because we were passing in "Lpackage/Class;" as the class name rather
than just "package/Class". Also, some classes have a weird situation
where some methods were moved around in later API versions. For example,
some put* and get* methods in Bundle were moved to BaseBundle in API 21.
If we only checked BaseBundle.put*, we would think they are API 21+
only. The workaround is to check both the top-level class and the
declaring class for a member, and choose the lower API level as the
minimal API level for that member.
This patch also fixes bugs in including the right class members.
For SDKProcessor we want to include all public members of a class,
including inherited members, because the private/protected members are
not part of the public API. For AnnotationProcessor, we want to include
all the members declared in that class, including private and
protected members, because we may want to access private/protected
members of our own Java classes from C++.
Trying to decipher MOZ_SUBCONFIGURE_ICU given its lack of indentation is
difficult at best. It looks like some lines have tabs, and those tabs
make everything line up right...convert everything to spaces to make
sure things line up correctly.
Some mozconfigs don't include mozconfig.linux*, and don't get gtk-related
definitions, so move them in a separate mozconfig. To avoid having two
files, one for 32-bit builds and one for 64-bit builds, rely on the
includer to set PKG_CONFIG_LIBDIR appropriately.
At the same time, move all the --enable-default-toolkit=cairo-gtk2 in that
new file in the case the gtk3 package wasn't pulled from tooltool.
There are a variety of ways that the parent and child process ensure that
the child process exits quickly in opt builds, but for AddressSanitizer
builds we want to let the child process to run to completion, so that we
can get a LeakSanitizer report.
This requires adding some addition LSan suppressions, because running
LSan in child processes detects some new leaks.
This will add more verbosity, with the intent of having the triggered
suppressions be displayed in the form:
"--23983-- used_suppression: 3 Bug 794372 /builds/slave/try-l64-valgrind-0000000000000/src/build/valgrind/cross-architecture.sup:90 suppressed: 20,736 bytes in 648 blocks"
It's been clear from user feedback that people don't realize that `mach
mercurial-setup` doesn't make any changes unless they tell it to.
Reinforce this message in the prompts printed by mach_boostrap.py.
Add a mozconfig fragment enabling rust on mac builds.
Include it in the official integration mozconfig files only.
Developer checkouts don't require rust.
This command is used by tab completion handlers. A user reported that
hitting tab in his shell resulted in the mercurial-setup out of date
check spewing output.
Having not configured or out-of-date tools benefits nobody. It slows
people down.
Version control tools are an integral part of working on Firefox. It is
important for version control tools to be configured optimally and to be
continuously updated so they stay optimal.
The `mach mercurial-setup` command exists to optimally configure
Mercurial for working on Firefox and other Mozilla projects.
This commit adds a pre-dispatch handler to mach that will verify
Mercurial is in a happy state. If `mach mercurial-setup` has never
executed, it will complain. If `mach mercurial-setup` hasn't been
executed in the past 31 days, it will complain.
Yes, aborting command execution and forcing people to context switch to
run `mach mercurial-setup` is annoying. First, we have carved out
several exceptions to this behavior, including detection for running in
automation, on the machines of curmudgeons, when Mercurial isn't being
used, and from non-interactive processes. Second, I argue that people
ignore optional notifications and that having persistently
poorly-configured tools is worse than a single context switch at most
every month. Therefore, the heavyhanded approach is justified.
In addition, if we did support a non-fatal notification, we would
introduce the problem of extra output from commands. If anyone was e.g.
parsing mach output, we could very likely break those systems. These
cases should be caught by the isatty() check or be running in a context
with MOZ_AUTOMATION set. But you never know.
Having not configured or out-of-date tools benefits nobody. It slows
people down.
Version control tools are an integral part of working on Firefox. It is
important for version control tools to be configured optimally and to be
continuously updated so they stay optimal.
The `mach mercurial-setup` command exists to optimally configure
Mercurial for working on Firefox and other Mozilla projects.
This commit adds a pre-dispatch handler to mach that will verify
Mercurial is in a happy state. If `mach mercurial-setup` has never
executed, it will complain. If `mach mercurial-setup` hasn't been
executed in the past 2 weeks, it will complain.
Yes, aborting command execution and forcing people to context switch to
run `mach mercurial-setup` is annoying. First, we have carved out
several exceptions to this behavior, including detection for running in
automation, on the machines of curmudgeons, when Mercurial isn't being
used, and from non-interactive processes. Second, I argue that people
ignore optional notifications and that having persistently
poorly-configured tools is worse than a single context switch at most
every 2 weeks. Therefore, the heavyhanded approach is justified.
In addition, if we did support a non-fatal notification, we would
introduce the problem of extra output from commands. If anyone was e.g.
parsing mach output, we could very likely break those systems. These
cases should be caught by the isatty() check or be running in a context
with MOZ_AUTOMATION set. But you never know.
A subsequent commit will want to access the state directory path without
possibly creating it. Make that possible by extracting path resolution
to its own function.
- Removed old robotium jar in replace for the newer one.
- Newer robotium has packaging changes which were updated in all tests.
- Updated in build.gradle and makefile.
Mozlog currently has two implementations. The top level package is based on the logging module and is
deprecated. The newer structured logging implementation lives in mozlog.structured. This patch swaps the
two, so the top level mozlog module contains the recommended implementation, while mozlog.unstructured
contains the old one.
We were de-referencing the checker variable after having moved it into
the array, which was causing a null pointer crash.
With this fixed, the plugin can be built with more recent versions of
clang.
Backed out changeset 58331e57de1c (bug 917999)
Backed out changeset 50f9123412c7 (bug 917999)
Backed out changeset 3b19643ec039 (bug 917999)
CLOSED TREE
Write a mozconfig fragment which makes the rust toolchain
provided by tooltool available for linux builds.
Use linux64 mozconfigs to enable rust for official builds of
that target. These aren't used outside of automation builds,
so including rust there will verify code on check-in without
requiring developers to install rust.
This removes ambiguity as to which modules are being imported, making
import slightly faster as Python doesn't need to test so many
directories for file presence.
All files should already be using absolute imports because mach command
modules aren't imported to the package they belong to: they instead
belong to the "mach" package. So relative imports shouldn't have been
used.
