#8274 closed patch (wontfix)
Make ScummVM compile with GCC 3.3.1
| Reported by: | SF/johneck | Owned by: | fingolfin |
|---|---|---|---|
| Priority: | normal | Component: | Port: Linux |
| Version: | Keywords: | ||
| Cc: | Game: |
Description
The latest CVS versions of ScummVM refuses to compile, it dies in all asserts() where pointers are used. Example: byte* tmp = some_function_method_or_other(); assert(tmp);
Will fail with "*byte is not of long int" (or similar). The way to fix this is to: byte* tmp = some_function_method_or_other(); assert(tmp != NULL);
And atleast it compiles with GCC 3.3.1 now. The patch fixes all these, atleast when trying to compile for the SDL backend.
Ticket imported from: #794066. Ticket imported from: patches/379.
Attachments (1)
Change History (13)
comment:1 by , 21 years ago
comment:2 by , 21 years ago
Strange. I distinctly remember adding the patch... Maybe Opera did something stupid.
Trying again.
comment:3 by , 21 years ago
This is the exact error:
g++ -Wp,-MMD,"scumm/.deps/actor.d",-MQ,"scumm/actor.o",-MP -O -Wall -Wuninitialized
-Wno-long-long -Wno-multichar -Wno-unknown-pragmas -g -ansi -W -Wno-unused-parameter
-pedantic -Wpointer-arith -Wcast-qual -Wcast-align -Wconversion -Wshadow -Wimplicit
-Wundef -Wnon-virtual-dtor -Wno-reorder -Wwrite-strings -fcheck-new -Wctor-dtor-privacy
-DHAVE_CONFIG_H -DUNIX -I. -Icommon -I/usr/include/SDL -D_REENTRANT -c scumm/actor.
cpp -o scumm/actor.o
scumm/actor.cpp: In member function `void Actor::animateCostume()':
scumm/actor.cpp:1038: error: invalid conversion from `byte*' to `long int'
scumm/actor.cpp: In member function `void Actor::animateLimb(int, int)':
scumm/actor.cpp:1070: error: invalid conversion from `byte*' to `long int'
make: *** [scumm/actor.o] Error 1
GCC version: Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/3.3.1/specs Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --enable-shared --enable-threads=posix --disable-checking --with-system-zlib --enable-__cxa_atexit --host=i386-redhat-linux Thread model: posix gcc version 3.3.1 20030811 (Red Hat Linux 3.3.1-1)
comment:4 by , 21 years ago
GCC 3.3.1 worked fine here and the debian prerelease of GCC 3.3.2 also works fine here
comment:5 by , 21 years ago
Hmm...
Might be a problem with the assert.h file then. That is provided by glibc. The glibc on this system is 2.3.2-71 (RedHat release). This is from Raw Hide (unstable, bleeding edge). I don't know exactly how much Red Hat patches glibc but this release has parts fetched from the glibc CVS.
And I don't know what ISO C says about assert but the glibc CVS might be botched or something. Or the CVS glibc may have found a bug in the assert code and fixed it. Not sure.
The assert does only have problems with non-POD types it seems.
comment:6 by , 21 years ago
I'm also running glibc 2.3.2-3 obviously the -version in both cases being the vendor release.
Once again this is the version found in debian sid and it works fine...
comment:7 by , 21 years ago
I have been playing around some more with this.
A small ugly test program: #include <stdio.h> #include <assert.h>
typedef unsigned char byte;
int main(void) { byte* nisse = NULL; assert(nisse);
return 0; }
compiling with:
gcc -Wall -o test test.c test.c: In function `main': test.c:9: warning: passing arg 1 of `__builtin_expect' makes integer from pointer without a cast
g++ -Wall -o test test.c test.c: In function `int main()': test.c:9: error: invalid conversion from `byte*' to `long int'
This is with the newer gcc-3.3.1-2 release from Red Hat.
Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/3.3.1/specs Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --enable-shared --enable-threads=posix --disable-checking --with-system-zlib --enable-__cxa_atexit --host=i386-redhat-linux Thread model: posix gcc version 3.3.1 20030814 (Red Hat Linux 3.3.1-2)
A small ugly test program, modified: #include <stdio.h> #include <assert.h>
typedef unsigned char byte;
int main(void) { byte* nisse = NULL; assert(nisse != NULL);
return 0; }
Works fine, both for gcc and g++, no warnings.
I am not sure what ISO C / ISO C++ says about assert(). Is this a Red Hat thing?
I did try to compile a bunch of other programs and so only ScummVM and ClanLib produced errors here (and in the case of ClanLib only two errors). The compiled stuff includes AbiWord, Wine, basic GNOME libraries, ...
comment:8 by , 21 years ago
ScummVM's use of assert() is 100% valid. Essentially, this assert(foo); is functionally equivalent to if (!(foo)) { ... /* do something */ } The error you get seems to imply something (compiler, C lib...) on your system is broken. The error would occur if for some reasons, assert on your system is not implemented as a macro, but rather (incorrectly!) as a function with a signature like void assert(long cond);
comment:9 by , 21 years ago
I know that assert(nisse); is valid C, and I am well aware of what assert do.
I did however figure out what broke it. It seems like RH decided to add __builtin_expect in the assert macro. This takes a long parameter. C manages but C++ will fail to make the cast from basically anything to long. The patch hides the problem since now __builtin_expect will be called with a boolean expression which can be cast without problems.
comment:10 by , 21 years ago
| Owner: | set to |
|---|---|
| Resolution: | → wontfix |
| Status: | new → closed |
comment:11 by , 21 years ago
I recommend you upgrade to a system with a non-broken compiler then. Sorry but it can't be our task to add work arounds for serious breakage some compiler vendor introduces into his compiler.
comment:12 by , 6 years ago
| Component: | → Port: Linux |
|---|
There is no patch attached.
BTW it seems this must be new in GCC 3.3.1, as things work just fine with 3.3 for me (and quite frankly, the error/warning makes no sense at all ... a compiler bug?)