mozglue/build/Makefile.in
author Jim Blandy <jimb@mozilla.com>
Tue, 24 Jun 2014 22:12:07 -0700
changeset 199513 69d61e42d5dfbf4588b72449249ff3e7f2125304
parent 177305 7d223c2e7fa9fe18c4380679df63cc9db41a63bc
child 205310 9565dc1c02c6a4917f1d8f68ba42654f4ee588c9
permissions -rw-r--r--
Bug 914753: Make Emacs file variable header lines correct, or at least consistent. DONTBUILD r=ehsan The -*- file variable lines -*- establish per-file settings that Emacs will pick up. This patch makes the following changes to those lines (and touches nothing else): - Never set the buffer's mode. Years ago, Emacs did not have a good JavaScript mode, so it made sense to use Java or C++ mode in .js files. However, Emacs has had js-mode for years now; it's perfectly serviceable, and is available and enabled by default in all major Emacs packagings. Selecting a mode in the -*- file variable line -*- is almost always the wrong thing to do anyway. It overrides Emacs's default choice, which is (now) reasonable; and even worse, it overrides settings the user might have made in their '.emacs' file for that file extension. It's only useful when there's something specific about that particular file that makes a particular mode appropriate. - Correctly propagate settings that establish the correct indentation level for this file: c-basic-offset and js2-basic-offset should be js-indent-level. Whatever value they're given should be preserved; different parts of our tree use different indentation styles. - We don't use tabs in Mozilla JS code. Always set indent-tabs-mode: nil. Remove tab-width: settings, at least in files that don't contain tab characters. - Remove js2-mode settings that belong in the user's .emacs file, like js2-skip-preprocessor-directives.

#
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.

DIST_INSTALL = 1

# For FORCE_SHARED_LIB
include $(topsrcdir)/config/config.mk

ifneq (1_1,$(MOZ_MEMORY)_$(or $(MOZ_NATIVE_JEMALLOC),$(FORCE_SHARED_LIB)))

ifneq (,$(filter WINNT,$(OS_ARCH)))
SDK_LIBRARY = $(IMPORT_LIBRARY)
else
ifdef FORCE_SHARED_LIB
SDK_LIBRARY = $(SHARED_LIBRARY)
else
SDK_LIBRARY = $(REAL_LIBRARY)
endif
endif

endif
 
MOZ_GLUE_LDFLAGS = # Don't link against ourselves

SHARED_LIBRARY_LIBS += $(call EXPAND_LIBNAME_PATH,mfbt,$(DEPTH)/mfbt)

ifneq (,$(ZLIB_IN_MOZGLUE)$(MOZ_LINKER))
ifdef MOZ_NATIVE_ZLIB
EXTRA_DSO_LDOPTS += $(MOZ_ZLIB_LIBS)
else
SHARED_LIBRARY_LIBS += $(MOZ_ZLIB_LIBS)
endif
endif

ifeq (WINNT,$(OS_TARGET))
mozglue.def: mozglue.def.in
	$(call py_action,preprocessor,$(if $(MOZ_REPLACE_MALLOC),-DMOZ_REPLACE_MALLOC) $(ACDEFINES) $< -o $@)

GARBAGE += mozglue.def

ifneq (,$(filter -DEFAULTLIB:mozcrt,$(MOZ_GLUE_LDFLAGS)))
# Don't install the import library if we use mozcrt
NO_INSTALL_IMPORT_LIBRARY = 1
endif


EXTRA_DSO_LDOPTS += \
	$(call EXPAND_LIBNAME,version) \
	$(NULL)

endif

ifeq (Darwin_1,$(OS_TARGET)_$(MOZ_REPLACE_MALLOC))
EXTRA_DSO_LDOPTS += \
  -Wl,-U,_replace_init \
  -Wl,-U,_replace_malloc \
  -Wl,-U,_replace_posix_memalign \
  -Wl,-U,_replace_aligned_alloc \
  -Wl,-U,_replace_calloc \
  -Wl,-U,_replace_realloc \
  -Wl,-U,_replace_free \
  -Wl,-U,_replace_memalign \
  -Wl,-U,_replace_valloc \
  -Wl,-U,_replace_malloc_usable_size \
  -Wl,-U,_replace_malloc_good_size \
  -Wl,-U,_replace_jemalloc_stats \
  -Wl,-U,_replace_jemalloc_purge_freed_pages \
  -Wl,-U,_replace_jemalloc_free_dirty_pages \
  $(NULL)

ifneq ($(MOZ_REPLACE_MALLOC_LINKAGE),compiler support)
EXTRA_DSO_LDOPTS += -flat_namespace
endif
ifeq ($(MOZ_REPLACE_MALLOC_LINKAGE),dummy library)
EXTRA_DSO_LDOPTS += -Wl,-weak_library,$(DEPTH)/memory/replace/dummy/$(DLL_PREFIX)replace_malloc$(DLL_SUFFIX)
endif
endif

ifeq (android, $(MOZ_WIDGET_TOOLKIT))
# To properly wrap jemalloc's pthread_atfork call.
EXTRA_DSO_LDOPTS += -Wl,--wrap=pthread_atfork
endif

ifdef MOZ_LINKER
ifeq (arm, $(TARGET_CPU))
EXTRA_DSO_LDOPTS += -Wl,-version-script,$(srcdir)/arm-eabi-filter
endif

endif

ifeq (Android, $(OS_TARGET))
WRAP_LDFLAGS := $(filter -Wl%,$(WRAP_LDFLAGS))
endif

include $(topsrcdir)/config/rules.mk

ifdef MOZ_MEMORY
ifeq (WINNT,$(OS_TARGET))
# Roll our own custom logic here for the import library

###############################################################################
#
# Linking Mozilla itself to jemalloc is not particularly difficult.  To do this
# we avoid linking directly to the Microsoft-provided CRT import libraries.
# Instead, we link to our own import library which we generate here.  To
# replace the CRT's malloc/free/other memory management symbols we export
# our own versions out of jemalloc.dll.  We then take the import library that
# the compiler generates for jemalloc.dll and combine it with the MS CRT import
# libraries.  We put our library on the command line first, and the CRT symbols
# are discarded in favor of our versions!
#
# Unfortunately that was too easy.  The CRT import library is not a standard
# import library that contains a list of symbols and whatnot.  It also includes
# object files that are linked into generated programs.  One of these,
# crtdll.obj is (as one might expect) linked into all DLLs that link against
# the CRT.  This file does things like run static C++ constructors when the
# DLL is attached, call DllMain, etc.
#
# In the CRT source all malloc/free calls are made to malloc_crt and free_crt.
# In debug builds these are both defined to malloc_dbg and free_dbg.  In opt
# builds malloc_crt is an actual function, implemented and exposed from the
# CRT.  free_crt is, however, defined to be just plain old free.  This works
# fine inside the CRT where malloc_crt and free operate on the same heap.
# Outside the CRT malloc_crt is in the CRT's heap, but free is in jemalloc's
# heap.  This causes much pain at shutdown :-(
#
# The obvious solution here is to override malloc_crt too.  Unfortunately,
# that doesn't work because the CRT expects to be able to call msize on this
# piece of memory deep inside the CRT, which will fail because it'll call the
# CRT's msize on a pointer in jemalloc's heap.
#
# Our solution to this is quite devious.  We take apart the CRT's import lib
# and remove the problematic object file.  We then poke at the object file's
# symbol table and replace '__imp__free' (which means grab free from some
# other DLL) with '__imp__frex'.  Then we define our own dummy no-op function
# in jemalloc.dll and export it as frex.  Then we put the CRT import lib
# back together with the patched crtdll.obj, glue it to the end of jemalloc's
# import library and link the rest of Mozilla to that.
#
# The result?  A binary that uses jemalloc, doesn't crash, and leaks a tiny
# amount of memory (32 words per DLL in the 2010 CRT) at shutdown.
#
###############################################################################

libs:: mozcrt.lib
	$(INSTALL) $(IFLAGS2) mozcrt.lib $(DIST)/lib

# And finally combine that with the jemalloc import library to get an import
# library that has our malloc/free/etc and the CRT's everything else
mozcrt.lib: $(IMPORT_LIBRARY) msvc_modified.lib
	lib -OUT:$@ $^

# Put the fixed object file back in
msvc_modified.lib: msvc_removed.lib crtdll_fixed.obj
	lib -OUT:$@ $^

# Fix the object file
crtdll_fixed.obj: crtdll.obj
	$(PYTHON) $(srcdir)/fixcrt.py

# Find the path of crtdll.obj
CRTDLL_FULLPATH=$(subst \,\\,$(shell lib -list msvc_combined.lib | grep crtdll\\.obj))

# Remove the broken object file, only after we have extracted it
msvc_removed.lib: msvc_combined.lib crtdll.obj
	lib -OUT:$@ msvc_combined.lib -REMOVE:$(CRTDLL_FULLPATH)

# Extract the broken object file out of the combined library
crtdll.obj: msvc_combined.lib
	lib -OUT:$@ $^ -EXTRACT:$(CRTDLL_FULLPATH)

# Grab both CRT libraries and combine them into one library to simplify things
msvc_combined.lib:
	lib -OUT:$@ $(WIN32_CRT_LIBS)
endif
endif # MOZ_MEMORY