Looking for user feedback

The following sample code shows a problem with the interaction between

GCC for SPARC 4.0.3 and a Solaris 10 header file. Ultimately, I believe

the fix is going to have to be in the header, changing

typedef struct queue {...} queue_t;

to

typedef struct stream_queue {...} queue_t;

or somesuch, which will be just fine as long as all users of this data

structure have been conscientious enough to use the typedef and not the

raw structure name. However, I do wonder why the Studio Express compiler

does not similarly object.

% which g++

/opt/gcc/bin/g++

% which CC

/opt/SPROexpress/opt/SUNWspro/bin/CC

% g++ --version

sparc-sun-solaris2.10-g++ (GCC) 4.0.3 (gccfss)

Copyright (C) 2006 Free Software Foundation, Inc.

This is free software; see the source for copying conditions. There is NO

warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

% cat queue.cc

#include <sys/socket.h>

#include <queue>

using std::queue;

% CC -c queue.cc

% g++ -c queue.cc

queue.cc:3: error: 'queue' is already declared in this scope

% # Now let's switch around the order of the lines in the file,

% # to see how that error arises ...

% vi queue.cc

% cat queue.cc

#include <queue>

using std::queue;

#include <sys/socket.h>

% CC -c queue.cc

% g++ -c queue.cc

/usr/include/sys/stream.h:67: error: template argument required for 'struct queue'

/usr/include/sys/stream.h:123: error: invalid type in declaration before ';' token

herteg at 2007-7-7 1:55:45 >

I'm seeing a hang (indefinite run time in cg) from a C file that used case ranges.

Here is a test program, reduced down from the file that had the error.

This is with a just downloaded version:

/opt/gcc/bin/gcc -v

Using built-in specs.

Target: sparc-sun-solaris2.10

Configured with: /net/tibia/export/bldmstr/nightly/20060817_mars_gcc.s10.opt.tarbuild/src/configure --prefix=/opt/gcc --enable-shared --with-system-zlib --enable-checking=release --disable-libmudflap --enable-languages=c,c++ --enable-version-specific-runtime-libs --with-gxx-include-dir=/opt/gcc/include/c++/4.0.3 --with-cpu=v9

Thread model: posix

gcc version 4.0.3 (gccfss)

Here is a script that will compile the test program with the default gcc

that comes with Solaris 10 in /opt/gcc/bin/gcc and with the gcc coupled

with the Sun Studio code generator. Of the three compiles in the test

script, the last one will hang.

#!/bin/ksh -x

# works

/usr/sfw/bin/gcc -c \

-DENABLE_case_dot_dot_dot \

show_case_dot_dot_dot_bug.c

# works

/opt/gcc/bin/gcc -c \

-UENABLE_case_dot_dot_dot \

show_case_dot_dot_dot_bug.c

# hangs

/opt/gcc/bin/gcc -c \

-DENABLE_case_dot_dot_dot \

show_case_dot_dot_dot_bug.c

And here is the test program, just one function that

has some case ranges in a switch statement. When

the case ranges are ifdef'ed out, the compile succeeds.

Test file name: show_case_dot_dot_dot_bug.c

typedef struct parameters {

int vectors;

int delay;

int comp;

int format;

int version;

} parameters;

int test(int val, parameters *s)

{

s->format = 200;

switch(val){

case 0x10000000:

s->version= 0;

s->vectors=0;

s->delay=1;

break;

case 0x10002000:

s->version= 3;

s->vectors=1;

s->delay=1;

s->comp=1;

break;

case 0x10003000:

s->version= 3;

s->vectors=1;

s->delay=1;

break;

case 0x10003001:

s->version= 3;

s->vectors=0;

s->delay=1;

break;

case 0x20001000:

#ifdef ENABLE_case_dot_dot_dot

case 0x20100000 ... 0x2019ffff:

#endif

s->delay=1;

break;

#ifdef ENABLE_case_dot_dot_dot

case 0x20200002 ... 0x202fffff:

#endif

case 0x30202002:

case 0x30203002:

s->delay=0;

break;

default:

s->delay=9;

}

return 0;

}

dave_p at 2007-7-7 1:55:45 >

Thank you for the quick reply. I worked around the problem by extracting

the case ranges out of the switch statement and into a if-else proceeding

the switch statement. I thought other people might run into this so it was

worth documenting the problem since they might not know to look for

case ranges in their code.

The testcase code I posted was extracted from one C file in a library of

audio and video encoders and decoders. The C file which had the

compile hang was preprocessed to remove all headers and then

iteratively ifdef'ed out until the function, and then the code, that

caused the compiler hang was located. The code was simplified

some more after that to keep the example small, so that and the

preprocessing might be why it looks like a synthetic test.

dave_p at 2007-7-7 1:55:45 >

> The upcoming patch/update will have a fix for this

> problem. We'll post an announcement when it's

> available on sdlc website. Hopefully in the next few

> weeks.

We've just released a patch for GCCfss 4.0.3. It's located at the same place place on SDLC website (the link is on the front page).

Please download it, remove old 4.0.3 packages and install new 4.0.3 packages.

This patch contains the fix for the above CG hang with large case ranges plus fixes for other problems reported on the forum.

alexey_ at 2007-7-7 1:55:46 >