| Message ID | 1630a93f8270ca090459be8cc7213221cc6250cf.1542060094.git.steadmon@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Makefile: use CXXFLAGS for linking fuzzers | expand |
On Mon, Nov 12, 2018 at 2:03 PM <steadmon@google.com> wrote: > > OSS-Fuzz requires C++-specific flags to link fuzzers. Passing these in > CFLAGS causes lots of build warnings. Using separate CXXFLAGS avoids > this. > That makes sense in this context, .... > CFLAGS = -g -O2 -Wall > +CXXFLAGS ?= $(CFLAGS) ... but out of context, just by reading the relevant part of the Makefile, a user might mistakenly assume we do some C++ trickery for standard compilation of Git. (Is that bad or do we just not care?) I wonder if setting the CXXFLAGS near or in the fuzz target would be better. > LDFLAGS = > ALL_CFLAGS = $(CPPFLAGS) $(CFLAGS) > ALL_LDFLAGS = $(LDFLAGS) > @@ -3098,14 +3099,14 @@ cover_db_html: cover_db > # An example command to build against libFuzzer from LLVM 4.0.0: > # > # make CC=clang CXX=clang++ \ > -# CFLAGS="-fsanitize-coverage=trace-pc-guard -fsanitize=address" \ > +# CXXFLAGS="-fsanitize-coverage=trace-pc-guard -fsanitize=address" \ > # LIB_FUZZING_ENGINE=/usr/lib/llvm-4.0/lib/libFuzzer.a \ > # fuzz-all > # > .PHONY: fuzz-all Maybe here? > > $(FUZZ_PROGRAMS): all > - $(QUIET_LINK)$(CXX) $(CFLAGS) $(LIB_OBJS) $(BUILTIN_OBJS) \ > + $(QUIET_LINK)$(CXX) $(CXXFLAGS) $(LIB_OBJS) $(BUILTIN_OBJS) \ > $(XDIFF_OBJS) $(EXTLIBS) git.o $@.o $(LIB_FUZZING_ENGINE) -o $@ Thanks, Stefan
steadmon@google.com writes: > OSS-Fuzz requires C++-specific flags to link fuzzers. Passing these in > CFLAGS causes lots of build warnings. Using separate CXXFLAGS avoids > this. We are not a C++ shop, so allow me to show ignorance about how projects that are OSS-Fuzz-enabled work. Do they use one set of CXXFLAGS when compiling the "real thing" and a separate set (perhaps one is subset of the other, or perhaps these two just have overlap) of CXXFLAGS when building to link with the fuzzer? What I am trying to get at is if this should be CXXFLAGS or CXX_FUZZER_FLAGS. If the OSS-Fuzz-enabled C++ projects use one set of flags for the "main" part of the project (to produce binaries to be run by the end users) and then link with extra flags to work with fuzzers, I would imagine that they won't call the latter CXXFLAGS but call it something else, and we probably should follow suit if that is the case. Not that we plan to (re)write the maint part of Git in C++ ever, so I am personally OK with sacrificing the most generic CXXFLAGS macro for the sole use of OSS-Fuzz linkage, but I'd prefer to leave the door open so that other things like OSS-Fuzz that require C++ can be added like your work does to the project. Thanks. > Signed-off-by: Josh Steadmon <steadmon@google.com> > --- > Makefile | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/Makefile b/Makefile > index bbfbb4292d..5462bc4b6b 100644 > --- a/Makefile > +++ b/Makefile > @@ -497,6 +497,7 @@ GIT-VERSION-FILE: FORCE > # CFLAGS and LDFLAGS are for the users to override from the command line. > > CFLAGS = -g -O2 -Wall > +CXXFLAGS ?= $(CFLAGS) > LDFLAGS = > ALL_CFLAGS = $(CPPFLAGS) $(CFLAGS) > ALL_LDFLAGS = $(LDFLAGS) > @@ -3098,14 +3099,14 @@ cover_db_html: cover_db > # An example command to build against libFuzzer from LLVM 4.0.0: > # > # make CC=clang CXX=clang++ \ > -# CFLAGS="-fsanitize-coverage=trace-pc-guard -fsanitize=address" \ > +# CXXFLAGS="-fsanitize-coverage=trace-pc-guard -fsanitize=address" \ > # LIB_FUZZING_ENGINE=/usr/lib/llvm-4.0/lib/libFuzzer.a \ > # fuzz-all > # > .PHONY: fuzz-all > > $(FUZZ_PROGRAMS): all > - $(QUIET_LINK)$(CXX) $(CFLAGS) $(LIB_OBJS) $(BUILTIN_OBJS) \ > + $(QUIET_LINK)$(CXX) $(CXXFLAGS) $(LIB_OBJS) $(BUILTIN_OBJS) \ > $(XDIFF_OBJS) $(EXTLIBS) git.o $@.o $(LIB_FUZZING_ENGINE) -o $@ > > fuzz-all: $(FUZZ_PROGRAMS)
On 2018.11.13 12:12, Junio C Hamano wrote: > steadmon@google.com writes: > > > OSS-Fuzz requires C++-specific flags to link fuzzers. Passing these in > > CFLAGS causes lots of build warnings. Using separate CXXFLAGS avoids > > this. > > We are not a C++ shop, so allow me to show ignorance about how > projects that are OSS-Fuzz-enabled work. Do they use one set of > CXXFLAGS when compiling the "real thing" and a separate set (perhaps > one is subset of the other, or perhaps these two just have overlap) > of CXXFLAGS when building to link with the fuzzer? > > What I am trying to get at is if this should be CXXFLAGS or > CXX_FUZZER_FLAGS. If the OSS-Fuzz-enabled C++ projects use one set > of flags for the "main" part of the project (to produce binaries to > be run by the end users) and then link with extra flags to work with > fuzzers, I would imagine that they won't call the latter CXXFLAGS > but call it something else, and we probably should follow suit if > that is the case. > > Not that we plan to (re)write the maint part of Git in C++ ever, so > I am personally OK with sacrificing the most generic CXXFLAGS macro > for the sole use of OSS-Fuzz linkage, but I'd prefer to leave the > door open so that other things like OSS-Fuzz that require C++ can be > added like your work does to the project. > > Thanks. OSS-Fuzz only provides one set of CXXFLAGS for use on both compiling project C++ project files as well linking the fuzzers themselves. So in the event that Git ever added any C++ sources, they would need to use the same set of CXXFLAGS. Given that, do you agree with Stefan that it is more intuitive to define CXXFLAGS next to the fuzzer build rules, since that's the only place it's used for now?
Josh Steadmon <steadmon@google.com> writes: > OSS-Fuzz only provides one set of CXXFLAGS for use on both compiling > project C++ project files as well linking the fuzzers themselves. So in > the event that Git ever added any C++ sources, they would need to use > the same set of CXXFLAGS. OK. > Given that, do you agree with Stefan that it is more intuitive to define > CXXFLAGS next to the fuzzer build rules, since that's the only place > it's used for now? I am not sure. Until we gain other C++ targets (in other words, while linking with fuzzer is the only consumer of CXXFLAGS), I'd consider it similar to SPARSE_FLAGS and SPATCH_FLAGS, i.e. settings specific to an auxiliary tool that supports our development process, and it would make more sense to define it near them higher in the Makefile. I'd probably feel differently if this were called FUZZ_CXXFLAGS or something like that, which would make its natural home next to the rule to build $(FUZZ_PROGRAMS), though.
diff --git a/Makefile b/Makefile index bbfbb4292d..5462bc4b6b 100644 --- a/Makefile +++ b/Makefile @@ -497,6 +497,7 @@ GIT-VERSION-FILE: FORCE # CFLAGS and LDFLAGS are for the users to override from the command line. CFLAGS = -g -O2 -Wall +CXXFLAGS ?= $(CFLAGS) LDFLAGS = ALL_CFLAGS = $(CPPFLAGS) $(CFLAGS) ALL_LDFLAGS = $(LDFLAGS) @@ -3098,14 +3099,14 @@ cover_db_html: cover_db # An example command to build against libFuzzer from LLVM 4.0.0: # # make CC=clang CXX=clang++ \ -# CFLAGS="-fsanitize-coverage=trace-pc-guard -fsanitize=address" \ +# CXXFLAGS="-fsanitize-coverage=trace-pc-guard -fsanitize=address" \ # LIB_FUZZING_ENGINE=/usr/lib/llvm-4.0/lib/libFuzzer.a \ # fuzz-all # .PHONY: fuzz-all $(FUZZ_PROGRAMS): all - $(QUIET_LINK)$(CXX) $(CFLAGS) $(LIB_OBJS) $(BUILTIN_OBJS) \ + $(QUIET_LINK)$(CXX) $(CXXFLAGS) $(LIB_OBJS) $(BUILTIN_OBJS) \ $(XDIFF_OBJS) $(EXTLIBS) git.o $@.o $(LIB_FUZZING_ENGINE) -o $@ fuzz-all: $(FUZZ_PROGRAMS)
OSS-Fuzz requires C++-specific flags to link fuzzers. Passing these in CFLAGS causes lots of build warnings. Using separate CXXFLAGS avoids this. Signed-off-by: Josh Steadmon <steadmon@google.com> --- Makefile | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)