Message ID | 20170125161417.31949-2-berrange@redhat.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 01/25/2017 10:14 AM, Daniel P. Berrange wrote: > If there are generated headers files to be used by the migration > code in '$BUILD_DIR/migration', these will not be found by the > relative include, an absolute include is needed instead. This > has not been a problem so far, since nothing has been generating > headers in sub-dirs, but the trace code will shortly be doing > that. So it is needed to list '-I$(BUILD_DIR)/$(@D)' as well as > '-I$(@D)' to ensure both directories are searched when building > target dependant code. So the search order ends up being: > > 1. source dir corresponding to input file (implicit by compiler) > 2. build dir corresponding to output file (absolute) > 3. build dir corresponding to output file (relative to cwd) > 4. top level build dir > 5. top level source dir > 6. top level source include/ dir Lots clearer than v3. > > One final complication is that the absolute '-I$(BUILD_DIR)/$(@D)' > will sometimes end up pointing to a non-existant directory if > that sub-dir does not have any target-independant files to be > built. Rather than try to dynamically filter this, a simple > 'mkdir' ensures $(BUILD_DIR)/$(@D) is guaranteed to exist at > all times. > > Signed-off-by: Daniel P. Berrange <berrange@redhat.com> > --- > rules.mak | 30 +++++++++++++++++++++++------- > 1 file changed, 23 insertions(+), 7 deletions(-) Reviewed-by: Eric Blake <eblake@redhat.com>
On Wed, Jan 25, 2017 at 04:14:10PM +0000, Daniel P. Berrange wrote: > One final complication is that the absolute '-I$(BUILD_DIR)/$(@D)' > will sometimes end up pointing to a non-existant directory if > that sub-dir does not have any target-independant files to be > built. Rather than try to dynamically filter this, a simple > 'mkdir' ensures $(BUILD_DIR)/$(@D) is guaranteed to exist at > all times. > > @@ -359,6 +374,7 @@ define unnest-vars > $(eval $(o:%.mo=%$(DSOSUF)): module-common.o $($o-objs)), > $(error $o added in $v but $o-objs is not set))) > $(shell mkdir -p ./ $(sort $(dir $($v)))) > + $(shell cd $(BUILD_DIR) && mkdir -p ./ $(sort $(dir $($v)))) > # Include all the .d files > $(eval -include $(patsubst %.o,%.d,$(patsubst %.mo,%.d,$($v)))) > $(eval $v := $(filter-out %/,$($v)))) After this change building QEMU leaves a lot of empty directories in the parent directory: $ mkdir empty_dir $ cd empty_dir $ git clone https://github.com/qemu/qemu $ cd qemu $ ./configure ... $ ls .. qemu $ make $ ls .. audio chardev fsdev linux-user net qom target backends crypto hw migration qapi replay ui block disas io nbd qemu slirp util Berto
On Sat, Feb 04, 2017 at 05:48:01PM +0200, Alberto Garcia wrote: > On Wed, Jan 25, 2017 at 04:14:10PM +0000, Daniel P. Berrange wrote: > > > One final complication is that the absolute '-I$(BUILD_DIR)/$(@D)' > > will sometimes end up pointing to a non-existant directory if > > that sub-dir does not have any target-independant files to be > > built. Rather than try to dynamically filter this, a simple > > 'mkdir' ensures $(BUILD_DIR)/$(@D) is guaranteed to exist at > > all times. > > > > @@ -359,6 +374,7 @@ define unnest-vars > > $(eval $(o:%.mo=%$(DSOSUF)): module-common.o $($o-objs)), > > $(error $o added in $v but $o-objs is not set))) > > $(shell mkdir -p ./ $(sort $(dir $($v)))) > > + $(shell cd $(BUILD_DIR) && mkdir -p ./ $(sort $(dir $($v)))) > > # Include all the .d files > > $(eval -include $(patsubst %.o,%.d,$(patsubst %.mo,%.d,$($v)))) > > $(eval $v := $(filter-out %/,$($v)))) > > After this change building QEMU leaves a lot of empty directories in > the parent directory: > > $ mkdir empty_dir > $ cd empty_dir > $ git clone https://github.com/qemu/qemu > $ cd qemu > $ ./configure ... > $ ls .. > qemu > $ make > $ ls .. > audio chardev fsdev linux-user net qom target > backends crypto hw migration qapi replay ui > block disas io nbd qemu slirp util Yuk, that's a horrible mistake, of course missed because git status won't tell you about stuff creatd in the parent of the repo :-( Regards, Daniel
diff --git a/rules.mak b/rules.mak index d5c516c..575a3af 100644 --- a/rules.mak +++ b/rules.mak @@ -26,8 +26,13 @@ QEMU_CXXFLAGS = -D__STDC_LIMIT_MACROS $(filter-out -Wstrict-prototypes -Wmissing # Flags for dependency generation QEMU_DGFLAGS += -MMD -MP -MT $@ -MF $(@D)/$(*F).d -# Same as -I$(SRC_PATH) -I., but for the nested source/object directories -QEMU_INCLUDES += -I$(<D) -I$(@D) +# Compiler searches the source file dir first, but in vpath builds +# we need to make it search the build dir too, before any other +# explicit search paths. There are two search locations in the build +# dir, one absolute and the other relative to the compiler working +# directory. These are the same for target-independent files, but +# different for target-dependent ones. +QEMU_LOCAL_INCLUDES = -I$(BUILD_DIR)/$(@D) -I$(@D) WL_U := -Wl,-u, find-symbols = $(if $1, $(sort $(shell $(NM) -P -g $1 | $2))) @@ -61,7 +66,9 @@ expand-objs = $(strip $(sort $(filter %.o,$1)) \ $(filter-out %.o %.mo,$1)) %.o: %.c - $(call quiet-command,$(CC) $(QEMU_INCLUDES) $(QEMU_CFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) -c -o $@ $<,"CC","$(TARGET_DIR)$@") + $(call quiet-command,$(CC) $(QEMU_LOCAL_INCLUDES) $(QEMU_INCLUDES) \ + $(QEMU_CFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) \ + -c -o $@ $<,"CC","$(TARGET_DIR)$@") %.o: %.rc $(call quiet-command,$(WINDRES) -I. -o $@ $<,"RC","$(TARGET_DIR)$@") @@ -74,16 +81,24 @@ LINK = $(call quiet-command, $(LINKPROG) $(QEMU_CFLAGS) $(CFLAGS) $(LDFLAGS) -o $(version-obj-y) $(call extract-libs,$1) $(LIBS),"LINK","$(TARGET_DIR)$@") %.o: %.S - $(call quiet-command,$(CCAS) $(QEMU_INCLUDES) $(QEMU_CFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) -c -o $@ $<,"CCAS","$(TARGET_DIR)$@") + $(call quiet-command,$(CCAS) $(QEMU_LOCAL_INCLUDES) $(QEMU_INCLUDES) \ + $(QEMU_CFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) \ + -c -o $@ $<,"CCAS","$(TARGET_DIR)$@") %.o: %.cc - $(call quiet-command,$(CXX) $(QEMU_INCLUDES) $(QEMU_CXXFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) -c -o $@ $<,"CXX","$(TARGET_DIR)$@") + $(call quiet-command,$(CXX) $(QEMU_LOCAL_INCLUDES) $(QEMU_INCLUDES) \ + $(QEMU_CXXFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) \ + -c -o $@ $<,"CXX","$(TARGET_DIR)$@") %.o: %.cpp - $(call quiet-command,$(CXX) $(QEMU_INCLUDES) $(QEMU_CXXFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) -c -o $@ $<,"CXX","$(TARGET_DIR)$@") + $(call quiet-command,$(CXX) $(QEMU_LOCAL_INCLUDES) $(QEMU_INCLUDES) \ + $(QEMU_CXXFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) \ + -c -o $@ $<,"CXX","$(TARGET_DIR)$@") %.o: %.m - $(call quiet-command,$(OBJCC) $(QEMU_INCLUDES) $(QEMU_CFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) -c -o $@ $<,"OBJC","$(TARGET_DIR)$@") + $(call quiet-command,$(OBJCC) $(QEMU_LOCAL_INCLUDES) $(QEMU_INCLUDES) \ + $(QEMU_CFLAGS) $(QEMU_DGFLAGS) $(CFLAGS) $($@-cflags) \ + -c -o $@ $<,"OBJC","$(TARGET_DIR)$@") %.o: %.dtrace $(call quiet-command,dtrace -o $@ -G -s $<,"GEN","$(TARGET_DIR)$@") @@ -359,6 +374,7 @@ define unnest-vars $(eval $(o:%.mo=%$(DSOSUF)): module-common.o $($o-objs)), $(error $o added in $v but $o-objs is not set))) $(shell mkdir -p ./ $(sort $(dir $($v)))) + $(shell cd $(BUILD_DIR) && mkdir -p ./ $(sort $(dir $($v)))) # Include all the .d files $(eval -include $(patsubst %.o,%.d,$(patsubst %.mo,%.d,$($v)))) $(eval $v := $(filter-out %/,$($v))))
Currently the search path is 1. source dir corresponding to input file (implicit by compiler) 2. top level build dir 3. top level source dir 4. top level source include/ dir 5. source dir corresponding to input file 6. build dir corresponding to output file Search item 5 is an effective no-op, since it duplicates item 1. When srcdir == builddir, item 6 also duplicates item 1, which causes a semantic difference between VPATH and non-VPATH builds. Thus to ensure consistent semantics we need item 6 to be present immediately after item 1. e.g. 1. source dir corresponding to input file (implicit by compiler) 2. build dir corresponding to output file 3. top level build dir 4. top level source dir 5. top level source include/ dir When srcdir == builddir, items 1 & 2 collapse into one, and items 3 & 4 collapse into one, but the overall search order is still consistent with srcdir != builddir A further complication is that while most of the source files are built with a current directory of $BUILD_DIR, target dependant files are built with a current directory of $BUILD_DIR/$TARGET. As a result, search item 2 resolves to a different location for target independant vs target dependant files. For example when building 'migration/ram.o', the use of '-I$(@D)' (which expands to '-Imigration') would not find '$BUILD_DIR/migration', but rather '$BUILD_DIR/$TARGET/migration'. If there are generated headers files to be used by the migration code in '$BUILD_DIR/migration', these will not be found by the relative include, an absolute include is needed instead. This has not been a problem so far, since nothing has been generating headers in sub-dirs, but the trace code will shortly be doing that. So it is needed to list '-I$(BUILD_DIR)/$(@D)' as well as '-I$(@D)' to ensure both directories are searched when building target dependant code. So the search order ends up being: 1. source dir corresponding to input file (implicit by compiler) 2. build dir corresponding to output file (absolute) 3. build dir corresponding to output file (relative to cwd) 4. top level build dir 5. top level source dir 6. top level source include/ dir One final complication is that the absolute '-I$(BUILD_DIR)/$(@D)' will sometimes end up pointing to a non-existant directory if that sub-dir does not have any target-independant files to be built. Rather than try to dynamically filter this, a simple 'mkdir' ensures $(BUILD_DIR)/$(@D) is guaranteed to exist at all times. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> --- rules.mak | 30 +++++++++++++++++++++++------- 1 file changed, 23 insertions(+), 7 deletions(-)