[CFT] Update to clang 3.4

[Dimitry Andric]

On 12 Jan 2014, at 13:12, Yamaya Takashi <yamayan at kbh.biglobe.ne.jp> wrote:
> buildworld is failed when WITH_LLDB=

Yes, this is a known issue. I discussed it with Ed Maste.  Clang 3.4
will have to be imported first, afterwards we can fix lldb.  


> some ports cannot build.
> 
> reason1: clang cannot handle some options.
> (libmad build)
> cc: error: unknown argument: '-fforce-addr'
> cc: error: unknown argument: '-fthread-jumps'
> cc: error: unknown argument: '-fcse-follow-jumps'
> cc: error: unknown argument: '-fcse-skip-blocks'
> cc: error: unknown argument: '-fregmove'
> cc: error: unknown argument: '-fschedule-insns2'
> (libtheora build)
> cc: error: unknown argument: '-fforce-addr'
> (poppler build)
> c++: error: unknown argument: '-fno-check-new'
> (py27-sqlite build)
> cc: error: unknown argument: '-R/usr/local/lib'
> (tbb build)
> c++: error: unknown argument: '-fno-schedule-insns2'
> (gstreamer-ffmpeg build)
> cc: error: unknown argument: '-fno-force-addr'

Indeed, this is likely the most troublesome aspect of the upgrade.

Most of the above options, like "-fcse-follow-jumps", "-fno-check-new",
etc are gcc-specific, and will never be supported by clang.  These
options will have to be removed, or made conditional on the compiler.

The -R option is not a compiler option, but an ld option, and is also
ambiguous: ld interprets it as --just-symbols when it is followed by the
name of a file, but as -rpath when it is followed by the name of a
directory.  In all cases, this should be replaced with -Wl,-rpath.


> reason2: c++ -std=c++11 detect bad c++ code which older clang cannot detect.
> (libproxy build)
> /usr/ports/net/libproxy/work/libproxy-0.4.6/libproxy/modules/wpad_dns_alias.cpp:30:23:
> error: cannot initialize return object of type 'libproxy::url *' with an
> rvalue of type 'bool'
>                if (lasturl) return false;
>                                    ^~~~~
> (liveMedia build)

I have not looked in detail at this one, but this looks like a simple
bug.  The code should not return a boolean when the return type of the
function is a libproxy::url pointer.  Should be easy to fix.

The new C++11 rules make it possible to avoid 'casting' a false value to
a NULL pointer, and clang 3.4 detects this.


> c++ -c -Iinclude -I../UsageEnvironment/include -I../groupsock/include
> -I. -DBSD=1 -DSOCKLEN_T=socklen_t -D_LARGEFILE_SOURCE=1
> -D_FILE_OFFSET_BITS=64 -DHAVE_SOCKADDR_LEN=1 -Wall -DBSD=1  -O2 -pipe
> -Qunused-arguments -march=native -fPIC -fno-strict-aliasing -std=c++11
> -Wno-c++11-narrowing -stdlib=libc++ -Wno-deprecated RTSPRegisterSender.cpp
> RTSPClient.cpp:916:25: error: comparison between pointer and integer
> ('const char *' and 'int')
>  if (&line[paramIndex] == '\0') return False; // the header is assumed
> to be bad if it has no parameters
>      ~~~~~~~~~~~~~~~~~ ^  ~~~~
> (mp4v2 build)

This looks like an incorrect comparison: a pointer is compared with a
literal character.  Should be easy to fix.


> c++ -DHAVE_CONFIG_H -I./include -I./include -I. -I. -Wall -Wformat -O2
> -pipe -Qunused-arguments -march=native -fno-strict-aliasing -std=c++11 
> -Wno-c++11-narrowing -stdlib=libc++ -fvisibility=hidden -c
> src/mp4container.cpp  -fPIC -DPIC -o src/.libs/mp4container.o
> src/mp4.cpp:679:20: error: cannot initialize return object of type
> 'mp4v2_ismacrypParams *' (aka 'mp4v2_ismacryp_session_params *') with an
> rvalue of type 'MP4TrackId' (aka 'unsigned int')
>            return MP4_INVALID_TRACK_ID;
>                   ^~~~~~~~~~~~~~~~~~~~
> ./include/mp4v2/general.h:45:33: note: expanded from macro
> 'MP4_INVALID_TRACK_ID'
> #define MP4_INVALID_TRACK_ID    ((MP4TrackId)0)       /**< Constant:
> invalid MP4TrackId. */
>                                ^~~~~~~~~~~~~~~
> (thunderbird build)

Again, an attempt is made to use an incorrect type for the return value
of a function.  Either the value should be explicitly cast to the
correct type (mp4v2_ismacrypParams *), or the MP4_INVALID_TRACK_ID
define should be fixed to be of the correct type.


> clang++ -o jsoptparse.o -c  -I../../../dist/system_wrappers_js -include
> /usr/ports/mail/thunderbird/work/comm-esr24/mozilla/js/src/config/gcc_hidden.h
> -DEXPORT_JS_API -DIMPL_MFBT -DMOZ_GLUE_IN_PROGRAM -DNO_NSPR_10_SUPPORT
> -I/usr/ports/mail/thunderbird/work/comm-esr24/mozilla/js/src -I..
> -I/usr/ports/mail/thunderbird/work/comm-esr24/mozilla/js/src/shell -I.
> -I../../../dist/include  -I/usr/local/include/nspr       -fPIC
> -Qunused-arguments -isystem/usr/local/include  -Qunused-arguments -Wall
> -Wpointer-arith -Woverloaded-virtual -Werror=return-type -Wtype-limits
> -Wempty-body -Werror=conversion-null -Wsign-compare
> -Wno-invalid-offsetof -Wno-c++0x-extensions -Wno-extended-offsetof
> -Wno-unknown-warning-option -Wno-return-type-c-linkage
> -Wno-mismatched-tags -O2 -pipe -Qunused-arguments -march=native -O3
> -fno-strict-aliasing -std=c++11 -Wno-c++11-narrowing -stdlib=libc++
> -fno-rtti -ffunction-sections -fdata-sections -fno-exceptions -pipe 
> -DNDEBUG -DTRIMMED -O2 -O3 -fomit-frame-pointer  -Qunused-arguments
> -isystem/usr/local/include  -DMOZILLA_CLIENT -include ../js-confdefs.h
> -MD -MP -MF .deps/jsoptparse.o.pp  /usr/ports/mail/thunderbir
> d/work/comm-esr24/mozilla/js/src/shell/jsoptparse.cpp
> /usr/ports/mail/thunderbird/work/comm-esr24/mozilla/js/src/shell/jsoptparse.cpp:256:22:
> error: comparison between pointer and integer ('char *' and 'int')
>        if (value[0] == '\0')
>            ~~~~~~~~ ^  ~~~~

Like the mp4v2 build, this is probably a simple bug.  The code should
not compare a pointer to an integer.  Most likely, the right side was
supposed to be NULL instead?

-Dimitry

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 203 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.freebsd.org/pipermail/freebsd-toolchain/attachments/20140112/183e1e6e/attachment.sig>