Created attachment 4666 [details]
preprocessModifiedFiles.ii.bz2

This is the preprocessor output, compressed with "bzip2 -9"

I can confirm this, the warnings are clearly bogus. Slightly changing the code makes 
the locations of the problem go away or move randomly. It takes someone with some 
more time than me right now to reduce this, though. 
 
One of the things that startle me is that it reports warnings in functions with the wrong 
name (i.e. it says the problem is in function "main", but the given line number is in 
another function). Maybe this has something to do with inlining then. 
 
W. 

I have found a smaller test case that shows the bug (down to 36 lines including
7 blank lines).  

I am attaching "bad1.i", a preprocessed C file that triggers the warning.

Here is the output from compiling bad1.i with gcc 3.3.1:
bilbo:~/JikesRVM/gccbug > cc --save-temps -v -Wunreachable-code -O3 -o bad1.S -S
bad1.i
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/specs
Configured with: ./configure --cache-file=config.cache
Thread model: posix
gcc version 3.3.1
 /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/cc1 -fpreprocessed bad1.i -quiet
-dumpbase bad1.i -auxbase-strip bad1.S -O3 -Wunreachable-code -version -o bad1.S
GNU C version 3.3.1 (i686-pc-linux-gnu)
        compiled by GNU C version 3.3.
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
bad1.c: In function `main':
bad1.c:13: warning: will never be executed
bilbo:~/JikesRVM/gccbug > 

The bug does not appear when I compile the very similar source file good1.c:

bilbo:~/JikesRVM/gccbug > cc --save-temps -v -Wunreachable-code -O3 -o good1.S
-S good1.i
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/specs
Configured with: ./configure --cache-file=config.cache
Thread model: posix
gcc version 3.3.1
 /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/cc1 -fpreprocessed good1.i
-quiet -dumpbase good1.i -auxbase-strip good1.S -O3 -Wunreachable-code -version
-o good1.S
GNU C version 3.3.1 (i686-pc-linux-gnu)
        compiled by GNU C version 3.3.
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
bilbo:~/JikesRVM/gccbug > 

Nor does it appear when I compile good2.i, which is bad1.i without the for loop:

bilbo:~/JikesRVM/gccbug > cc --save-temps -v -Wunreachable-code -O3 -o good2.S
-S good2.i
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/specs
Configured with: ./configure --cache-file=config.cache
Thread model: posix
gcc version 3.3.1
 /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/cc1 -fpreprocessed good2.i
-quiet -dumpbase good2.i -auxbase-strip good2.S -O3 -Wunreachable-code -version
-o good2.S
GNU C version 3.3.1 (i686-pc-linux-gnu)
        compiled by GNU C version 3.3.
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
bilbo:~/JikesRVM/gccbug > 

Created attachment 4673 [details]
bad1.i: preprocessed C that triggers the problem.

 preprocessed C that triggers the problem.

Here is the output from compiling bad1.i with gcc 3.3.1:
bilbo:~/JikesRVM/gccbug > cc --save-temps -v -Wunreachable-code -O3 -o bad1.S
-S bad1.i
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/specs
Configured with: ./configure --cache-file=config.cache
Thread model: posix
gcc version 3.3.1
 /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/cc1 -fpreprocessed bad1.i
-quiet -dumpbase bad1.i -auxbase-strip bad1.S -O3 -Wunreachable-code -version
-o bad1.S
GNU C version 3.3.1 (i686-pc-linux-gnu)
	compiled by GNU C version 3.3.
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
bad1.c: In function `main':
bad1.c:13: warning: will never be executed
bilbo:~/JikesRVM/gccbug >

Created attachment 4674 [details]
good1.i, one-line change that makes the problem go away

good1.i, one-line change that makes the problem go away


bilbo:~/JikesRVM/gccbug > cc --save-temps -v -Wunreachable-code -O3 -o good1.S
-S good1.i
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/specs
Configured with: ./configure --cache-file=config.cache
Thread model: posix
gcc version 3.3.1
 /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/cc1 -fpreprocessed good1.i
-quiet -dumpbase good1.i -auxbase-strip good1.S -O3 -Wunreachable-code -version
-o good1.S
GNU C version 3.3.1 (i686-pc-linux-gnu)
	compiled by GNU C version 3.3.
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
bilbo:~/JikesRVM/gccbug >

Created attachment 4675 [details]
good2.i, preprocessed C source that doesn't show the problem.

The bug doesn't appear when I compile good2.i, which is bad1.i without the for
loop:

bilbo:~/JikesRVM/gccbug > cc --save-temps -v -Wunreachable-code -O3 -o good2.S
-S good2.i
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/specs
Configured with: ./configure --cache-file=config.cache
Thread model: posix
gcc version 3.3.1
 /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/cc1 -fpreprocessed good2.i
-quiet -dumpbase good2.i -auxbase-strip good2.S -O3 -Wunreachable-code -version
-o good2.S
GNU C version 3.3.1 (i686-pc-linux-gnu)
	compiled by GNU C version 3.3.
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
bilbo:~/JikesRVM/gccbug > 
bilbo:~/JikesRVM/gccbug >

bad1.c is a nice example, but you didn't go all the way :-) Here's something even smaller: 
------------------------------------------- 
int g(); 
void f (int i) { 
  for (; g();) 
    if (g()) { 
      ++i; 
      break; 
    } 
} 
------------------------------------ 
tmp/g> ~/bin/gcc-3.3/bin/gcc -Wunreachable-code -O3 -c x.c 
x.c: In function `f': 
x.c:5: warning: will never be executed 
 
That's clearly bogus. The same happens with present mainline, with an additional 
twist: there's an empty line sneaking into the message: 
tmp/g> ../build-gcc/gcc-install/bin/gcc -Wunreachable-code -O3 -c x.c 
x.c: In function `f': 
 
x.c:5: warning: will never be executed 
 
Thanks fro reducing your testcase! 
W. 

Forgot to say this -- it's a regression in 3.3 and mailine w.r.t. 3.2. 
 
W. 

Created attachment 4676 [details]
testcase (11 lines)

The warning appears with this testcase (11 lines of preprocessed C).
bilbo:~/JikesRVM/gccbug > gcc -O3 -Wunreachable-code -v --save-temps bad2.i -S
-o bad2.S
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/specs
Configured with: ./configure --cache-file=config.cache
Thread model: posix
gcc version 3.3.1
 /usr/local/lib/gcc-lib/i686-pc-linux-gnu/3.3.1/cc1 -fpreprocessed bad2.i
-quiet -dumpbase bad2.i -auxbase-strip bad2.S -O3 -Wunreachable-code -version
-o bad2.S
GNU C version 3.3.1 (i686-pc-linux-gnu)
	compiled by GNU C version 3.3.
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
bad2.i: In function `main':
bad2.i:6: warning: will never be executed
bilbo:~/JikesRVM/gccbug >

Wolfgang Bangerth's test case in Comment #7 is superior to any of the attached
ones that I wrote.  (I added comment #9 unaware that Wolfgang had added #7 and #8).

The regression in PR 12082 was introduced or exposed by this patch:

--- gcc/gcc/ChangeLog ---

Don Feb 28 11:07:36 CET 2002  Jan Hubicka  <jh@suse.cz>

        * basic-block.h (BB_REACHABLE): Renumber.
        (BB_DIRTY, BB_NEW): New flags.
        (clear_bb_flags): Declare.
        (update_life_info_in_dirty_blocks): Declare.
        * cfg.c (clear_bb_flags): New function.
        * cfgrtl.c (create_basic_block_structure): Set flags to BB_NEW.
        * emit-rtl.c (add_insn_after, add_insn_before, remove_insn,
        reorder_insns, emit_insn_after): Mark block as dirty.
        * flow.c (update_life_info): Fix clearing of PROP_LOG_LINKS.
        (update_life_info_in_dirty_blocks): New function.
        * recog.c (apply_change_group): Dirtify block.

        * cse.c (cse_insn): Reorder emitting of jump insn to keep
        cfg consistent.
        * gcse.c (delete_null_pointer_checks): Likewise.

        * toplev.c (dump_file_index): Move cse2 after bp,
        add DFI_null
        (dump_file_info): Similary.
        (rest_of_compilation): Avoid most of CFG rebuilds;
        do first if converision after null pointer checks, do cse2
        after branch prediction; avoid full liveness rebuild after
        initializing subregs.
        * invoke.texi (-d options): Document -du, renumber.

        * cfgcleanup.c (bb_flags): Remove BB_UPDATE_LIFE.
        (notice_new_block): Do not set BB_UPDATE_LIFE.
        (try_forward_edges, merge_blocks_move_predecessor_nojumps,
         merge_blocks_move_successor_nojumps, merge_blocks,
         try_crossjump_to_edge): Likewise.
        (try_optimize_cfg): Likewise; use update_life_info_in_dirty_blocks.
        * cfgrtl.c (merge_blocks_nomove): Copy b's flags to a.
        * ifcvt.c (SET_UPDATE_LIFE, UPDATE_LIFE): Kill.
        (merge_of_block): Do not use life_data_ok.
        (find_if_case_1): Do not use SET_UPDATE_LIFE.
        (if_convert): Use BB_DIRTY mechanizm to update life.
        * lcm.c (optimize_mode_switching): Update
        update_life_info_in_dirty_blocks

The regression hunt took place on i686-pc-linux-gnu using Wolfgang's
small test case from comment #7 compiled with "-O3 -Wunreachable-code".

Subject: Bug 12082

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_3-branch
Changes by:	hubicka@gcc.gnu.org	2003-09-06 21:26:52

Modified files:
	gcc            : ChangeLog calls.c cfgcleanup.c 

Log message:
	PR target/12070
	* calls.c (emit_library_call_value_1): Fix saving of BLKmode arguments.
	
	PR opt/12082
	* cfgcleanup.c (try_simplify_condjump): Avoid unreachable code warning.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.16114.2.724&r2=1.16114.2.725
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/calls.c.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.244.2.10&r2=1.244.2.11
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cfgcleanup.c.diff?cvsroot=gcc&only_with_tag=gcc-3_3-branch&r1=1.68.2.5&r2=1.68.2.6

Subject: Bug 12082

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	hubicka@gcc.gnu.org	2003-09-06 21:50:22

Modified files:
	gcc            : ChangeLog calls.c cfgcleanup.c 

Log message:
	PR target/12070
	* calls.c (emit_library_call_value_1): Fix saving of BLKmode arguments.
	
	PR opt/12082
	* cfgcleanup.c (try_simplify_condjump): Avoid unreachable code warning.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.988&r2=2.989
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/calls.c.diff?cvsroot=gcc&r1=1.297&r2=1.298
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cfgcleanup.c.diff?cvsroot=gcc&r1=1.91&r2=1.92

Fixed by the patch above for 3.3.2 and 3.4.