Bugs introduced in GCC?
Colin Watson
cjwatson at ubuntu.com
Wed Sep 3 16:26:09 UTC 2008
On Tue, Sep 02, 2008 at 10:41:49PM -0400, Richard M. Stallman wrote:
> In terms of actual volume of patches against upstream, and excluding SVN
> updates, the diff from the base Debian 4.2.3-2 version and Ubuntu 8.04's
> 4.2.3-2ubuntu7 is about 40 KB.
>
> > If so, what is it meant to change?
>
> Matthias (CCed) should be able to give more information on any
> particular patches that are of concern.
>
> I didn't even imagine trying to read them myself!
My apologies for being vague; I wanted to reply to you, but didn't have
anything like a reasonable summary of all the Debian/Ubuntu changes to
hand. It sounded like your direct concerns were the apparent volume and
this specific bug, anyway.
However, I've just found the summary, which is in
/usr/share/doc/gcc-4.2-base/README.Debian.gz in the built binary
packages. It follows for your convenience in case you're interested.
Patches that Ubuntu applied in this version:
svn-updates:
updates from the 4.2 branch upto 20080329.
svn-doc-updates:
updates from the 4.2 branch upto 20080308 (documentation).
rename-info-files:
Allow transformations on info file names. Reference the
transformed info file names in the texinfo files.
gcc-directives-only-doc:
Backport from trunk: -fdirectives-only (when preprocessing,
handle directives, but do not expand macros).
documentation changes
sparc-niagara2-doc:
Add niagara2 optimization support
i386-biarch-doc:
biarch patches for i386/x86_64
gcc-version:
Add package identification to the gcc version string
gcc-textdomain:
Set gettext's domain and textdomain to the versioned package name.
gcc-driver-extra-langs:
Add options and specs for languages that are not built from a source
(but built from separate sources).
gcc-hash-style-both:
Link using --hash-style=both (alpha, amd64, ia64, i386, powerpc, ppc64, s390, sparc)
libstdc++-pic:
Build and install libstdc++_pic.a library.
libstdc++-doclink:
adjust hrefs to point to the local documentation
libstdc++-doxygen:
libstdc++ doxygen docs: Set SHORT_NAMES to YES
libjava-stacktrace:
libgcj: Lookup source file name and line number in separated
debug files found in /usr/lib/debug
libjava-subdir:
- Set the libjava sublibdir to /usr/lib/gcj-4.2
- Set the default libgcj database dir to /var/lib/gcj-4.2
libjava-jnipath:
- Add /usr/lib/jni to java.library.path.
- When running the i386 binaries on amd64, look in
- /usr/lib32/gcj-x.y and /usr/lib32/jni instead.
libjava-sjlj:
Don't try to use _Unwind_Backtrace on SJLJ targets.
See bug #387875, #388505, GCC PR 29206.
libjava-rpath:
- Link ecjx and gij with -rpath $(dbexecdir)
libjava-jar:
gjar: support @ arguments.
libjava-nojavac:
libjava/classpath: Don't require javac without java maintainer mode.
libjava-armel-ldflags:
Adjust libjava extra_ldflags for armel.
libjava-xulrunner1.9:
Add configure check for xulrunner-1.9.
libffi-configure:
Add --enable-libffi option to toplevel configure script
pr20218:
Backport from mainline for PR middle-end/17982, PR middle-end/20218.
pr20218-mips:
Backport from mainline for PR middle-end/20218, mips part.
pr28102:
Apply proposed fix for PR target/28102.
pr22244:
Improve debug info for packed arrays with constant bounds (PR fortran/22244)
pr30961:
<your description>
libgcc-tramp-ppc32:
Fix libgcc.a(tramp.o) on ppc32
gcc-directives-only:
Backport from trunk: -fdirectives-only (when preprocessing,
handle directives, but do not expand macros).
sparc-niagara2:
Add niagara2 optimization support
alpha-no-ev4-directive:
boehm-gc-nocheck:
Disable running the boehm-gc testsuite. Hangs the buildd at least on hppa.
boehm-gc-getnprocs:
boehm-gc/pthread_support.c (GC_get_nprocs): Use sysconf as fallback.
note-gnu-stack:
Add .note.GNU-stack sections for gcc's crt files, libffi and boehm-gc
Taken from FC.
arm-pr28516:
Fix PR28516 on ARM
arm-pr30486:
Fix 30486 on ARM
arm-unbreak-eabi-armv4t:
Fix armv4t build on ARM
hurd-profiling:
Fix profiling on hurd-i386.
hurd-fast-math:
Support -ffast-math on hurd-i386.
m68k-fjump:
gcc/config/m68k/m68k.md:
always use as fjcc pseudo op, we rely heavily on as
to generate the right size for the jump instructions
fixes #359281
m68k-save_pic:
gcc/config/m68k/m68k.c:
correctly save the pic register, when not done by reload()
(fixes _Unwind_RaiseException and thus exception handling)
fixes 345574
m68k-dwarf:
correct the dwarf frame information, but preserve compatibility
m68k-split_shift:
use correct predicates for long long shifts and use more splits
fixes 381572
m68k-limit_reload:
correctly limit reload class
fixes 375522
m68k-prevent-qipush:
prevent combine from creating a byte push on the stack (invalid on m68k)
m68k-return:
don't use single return if fp register have to be restored
fixes 386864
m68k-jumptable:
Don't force byte offset when accessing the jumptable, gas can
generate the correct offset size instead.
m68k-sig-unwind:
add support for unwinding over signal frames
m68k-peephole:
Convert some text peepholes to rtl peepholes, so the correct
DWARF2 information can be generated for stack manipulations.
(Keep a few peepholes temporarily disabled.)
m68k-align-code:
use "move.l %a4,%a4" to advance within code.
m68k-align-stack:
try to keep the stack word aligned
m68k-secondary-addr-reload:
add secondary reloads to allow reload to get byte values into addr regs
m68k-notice-move:
don't set cc_status for fp move without fp register
m68k-fpcompare:
bring fp compare early to its desired form to relieve reload,
allow any fp constant during and after reload (even special constants)
in case reload can't find a free register for them.
m68k-symbolic-operand:
better recognize symbolic operands in addresses
m68k-bitfield-offset:
only use constant offset for register bitfields
(combine expects shifts, but does a rotate)
mips-libspec:
mips*-linux: Always imply -lpthread for -pthread
pr24170:
Fix PR libgcj/24170.
pr32889:
Proposed patch to fix root cause for PR middle-end/32889.
pr28322:
Fix PR other/28322, GCC new warnings and compatibility.
libobjc-gc-link:
Link libobjc_gc with libgcjgc_convenience.la.
gcc-ssp-default:
Turn on -fstack-protector by default for C, C++, ObjC, ObjC++.
Build libgcc using -fno-stack-protector.
deb-protoize:
build protoize/unprotoize binaries
gcc-d-lang:
Add D options and specs for the gcc driver.
cpu-default-i486:
set default 32bit ix86 architecture to i486
reporting:
Add Debian URL for bug reporting instructions.
multiarch-include:
biarch-include.dpatch
Adds biarch include directories
/usr/local/include/<arch>-linux-gnu
/usr/include/<arch>-linux-gnu
to the system include paths, depending on 32/64 bit mode.
gcc-ice-hack:
2004-01-23 Jakub Jelinek <jakub at redhat.com>
* system.h (ICE_EXIT_CODE): Define.
* gcc.c (execute): Don't free first string early, but at the end
of the function. Call retry_ice if compiler exited with
ICE_EXIT_CODE.
(retry_ice): New function.
* diagnostic.c (diagnostic_count_diagnostic,
diagnostic_action_after_output, error_recursion): Exit with
ICE_EXIT_CODE instead of FATAL_EXIT_CODE.
libjava-disable-static:
Disable building the static libjava.
gcc-ice-apport:
Report an ICE to apport (if apport is available
and the environment variable GCC_NOAPPORT is not set)
libjava-nobiarch-check:
For biarch builds, disable the testsuite for the non-default architecture
for runtime libraries, which are not built by default (libjava).
config-ml:
disable some biarch libraries for biarch builds
i386-biarch:
biarch patches for i386/x86_64
disable-configure-run-check:
don't bogusly check if the 64bit binaries also work,
which does break on 32bit kernels.
> It's just that the added bug (which I forwarded) made me worry.
> I'm glad you are working on fixing this, and I hope you will
> make a new release.
Yes; we're debating its exact content, but the fix for this particular
bug is uncontroversial.
Thanks,
--
Colin Watson [cjwatson at ubuntu.com]
More information about the Ubuntu-devel-discuss
mailing list