f78 Issues from previous releases Contents | Index | Previous | Next

Issues from previous releases

The following documentation discusses issues from previous releases. These issues can provide clues that may help if you run into problems with the current release.

General issues with the tools
C and C++ compiler issues
Debugger issues
Assembler issues
Linker issues
Rebuilding issues

General issues with the tools

No tapes or floppy disks shipped by default. The installation media is now CD format.
None of the libraries were thread-safe.
Some tools were not available for the following platform environments.
GNUPro Toolkit did not include a linker (ld) for the DEC Alpha running Digital Unix (formerly OSF/1). The native linker was the default.
GNUPro Toolkit required the SGI Irix operating system's C library and include files in a native configuration. The SGI Irix operating system does not contain these files by default, but they are included in a separate developer's package. You cannot use the GNUPro Toolkit without this package.
GNU ld was not included for the HP 9000/700 native in the native configuration.
If you ran dejagnu in an Emacs shell buffer, expect generated incorrect results for pass-fail.
Finding a shared library at run-time was a problem with the (default) C++ library, libstdc++, as well as with libg++.

ffb

Added '/usr/progressive/lib' to LD_LIBRARY_PATH environment variable.
Used the following appropriate option when linking:

For Solaris:

...-R/usr/progressive/lib...

For DEC Unix/OSF1 and Irix 5:

...-Wl, -rpath,/usr/progressive/lib...

Removed or renamed completely the shared libraries under the installation directory ('/usr/progressive/lib/libg++.s* ' and '/usr/progressive/lib/libstdc++.s* ').
Linked with -static to avoid using any shared libraries.
The library directory, '/usr/progressive/lib ', was different if GNUPro Toolkit was installed in a non-standard location. This directory was thought of as '$GCC_EXEC_PREFIX/../ ' if using 'GCC_EXEC_PREFIX '.
After rebuilding from source, in order to report problems, a user reran install-sid in order to reset customer-id in send-pr . Previously, the default value was the value in the send-pr program in a path at the time of configuration.

The following issues were pertinent to the PowerPC environment.

Added '/usr/progressive/lib' to LD_LIBRARY_PATH environment variable.
-mno-fused-madd was a new option, preventing the compiler from combining a floating point multiply with an addition or subtraction. Generally, the compiler would do this optimization, if it was important to get the exact rounding as specified by the IEEE 754 floating point standard (on the Power and PowerPC machines, the combined multiply and add/subtract instruction does not do a round operation between the multiply and the addition/subtraction).
The -mcpu=403 switch did a -mstrict-align operation, since the 403 had no support for unaligned memory operations.
With the previous release, better code was generated when adding or doing logical operations with large constants.
Two new switches, -mads and -myellowknife, were added to the powerpc-eabi toolchain.
The -mads switch links in the appropriate libraries for the Motorola ADS target boards (860/821/823).
-myellowknife, the other switch, links in the appropriate libraries for the Motorola Yellowknife target boards (603e/604e).
Jump tables, for switch statements, were in the .rodata section, and not the .text ffb section, so it wasn't necessary to include the text section in the data address mapping.
wchar_t became 4 bytes instead of 2 (and "a"l -strings consisted of 4 byte characters instead of 2).
The -mregnames switch was only passed to the assembler for assembling .s and .S files (i.e., it was not passed to the assembler when assembling the output of the compiler). This is so that global symbols like "r0" in C code wouldn't get confused with register names.
PowerPC configuration included a simulator contributed by Andrew Cagney.

The following support enhancements were made to both the powerpc-aix and powerpc-eabi configurations.

Support was added for -mcpu=505,-mcpu=602, -mcpu=860, -mcpu=821, and -mcpu=power2.
The default PowerPC model is now 604, not 601.
-mtune=xxx was added to select scheduling parameters such as -mcpu=xxx, but not to select use of cpu specific instructions.
If configuring and building the compiler, the switch, --with-cpu=xxx, will allow for selecting the default processor.
Instruction timings had been improved.
The compiler was set to generate correct code for a nor instruction combined with a compare instruction.

__attribute__((longcall)) was added to function attributes so that the function is now always called through a pointer. This allows the function to be located anywhere in program memory.

The following enhancements were made only to the powerpc-eabi support.

The default PowerPC model is now 604, not 601.
A new switch, -msdata was added to put small static and global items in small data regions. This allows them to be referenced with 1 instruction instead of 2.
long long was set to pass according to the System V/EABI specifications, (in other words, they always pass in odd/even register pairs, never even/odd).
Code was changed for the functions prolog and epilog so that references beyond the end of the current stack pointer never generate.
The assembler added support for generating negative address, GOT, PLT, and small data, relative to section start relocations.
The linker added support for negative address, GOT, and small data, relative to section start relocations. PLT relocations were not handled.
An enhancement was made to the powerpc-aix support. The default for AIX 4.1 was changed to -mcpu=common. This replaced the default of the machine on which the compiler was configured (either Power or PowerPC)
The PowerPC simulator must be built with the gcc compiler. It uses several gcc extensions, as well as requiring an ISO standa ffb rd compiler. If you configure your powerpc-eabi build without using gcc, it will not build the simulator.

The following issues are pertinent to the m68k environments.

The m68k-aout and m68k-coff toolchains now include support for converting executables files into IEEE-695 format. The IEEE-695 format is used by some emulators.
To convert a file, first link your program into a fully linked executable. Then run the following script.

m68k-coff-objcopy -O ieee --debugging file file.x

m68k-coff

file

file.x

To use the m68k-aout toolchain, m68k-aout-objcopy is the corresponding tool.
The --debugging option directs objcopy to translate debugging information into the IEEE-695 format. This option only works if you use the stabs debugging format. The stabsdebugging format is the default for the m68k-aout toolchain.
For the m68k-coff toolchain, you must compile your files using the option, -gstabs+ , rather than a simple -g, in order to get stabs debugging information.
A conversion facility is available for other toolchains, but the binaries built by Cygnus do not provide it. To use it, you must rebuild the programs from source, as described in Rebuilding From Source in GNUPro Advanced Topics.
gcc has an option, '-m68332', to be used specifically when compiling for the Motorola 68332 board. ( gcc also has an updated option, -m68302. The 68302 technically isn't a cpu 32 chip.)
Running the configure script, add '--enable-targets=CPU-ieee', where CPU is the processor type (such as i960). Cygnus had not tested the convertor for processors other than the m68k .
The Motorola cpu 32 and cpu 32+ targets are part of the family of 68000 chips, which Cygnus supports. There are a few options to help you compile code for these targets.

gcc

m68332-aout

m68332-coff

-m68332

GNU as accepts the following board-specific options: -mcpu32, -m68331, -m68332, -m68333, -m68340, and -m68302.

Contact Cygnus for more information on our support for cpu 32 and cpu 32+ targets.

C and C++ compiler issues

gcc

g++

Member function templates have support.

    struct S {

      template <class T> operator T();

      /* ... */

};

    template <class T>

    S::operator T()

      /* ... */

    main ()

      S s;

      int i = s;

      void *p = s;

Explicit qualification of function templates has support.

template <class T, class U> T implicit_cast (U u) { return u; }

int i = implicit_cast<int>(1.5);

-fguiding-decls

   template <class T> struct A {

      friend ostream& operator<< (ostream &, const A &);

      /* ... */

};

    template <class T> ostream&

    operator<< (ostream &o, const A<T>& a)

      /* ... */

    main ()

      A<int> a;

      cout << a;

Two exception handling mechanisms are used.

One, which is supported on all platforms, uses setjmp and longjmp to unwind stack frames and slow down your code.

The second mechanism uses DWARF2-format data to unwind stack frames, and has no time impact on code that doesn't throw exceptions, but does require that the unwind info be written out. The DWARF2-unwinder is the default on systems where it is supported; setjmp/longjmp is the default ev ffb erywhere else.

To select between the two, use -fsjlj-exceptions , except, where the DWARF2 unwinder is not supported, the old mechanism, EH , will be used. Turn EH off with the flag, -fno-exceptions .

Standard usage syntax for the std namespace is supported; std is treated as an alias for global scope. General namespaces are still not supported.
The flag, -Wno-pmf-conversion, tells the compiler not to warn about converting from a bound member function pointer to function pointer.
A flag, -Weffc++, is used for violations of some of the style guidelines in the "Effective C++" books by Scott Meyers.
__FUNCTION__ and __PRETTY_FUNCTION__ are treated as variables by the parser; previously they were treated as string constants.
Code like printf (__FUNCTION__ ": foo") must be rewritten to be:

 printf ("%s: foo", __FUNCTION__)

Local static variables in extern inline functions will be shared between translation units.
-fvtable-thunks is supported for all targets, and is the default for Linux with glibc 2.x (also called libc 6.x).
A new flag, -Wold-style-cast, can be used to warn if an old-style (C-style) cast is used within a C++ program.

hppa*-*-proelf (or a HP PRO target) no longer includes floating point support by default. Therefore, it is no longer necessary to include -msoft-float on either the compilation or link line for these targets.
A public review copy of the December 1996 Draft of the ANSI C++ Standard is available. For PostScript and PDF (using Adobe Acrobat) versions, see the archives: ftp://research.att.com/dist/c++std/WP. For HTML and ASCII versions, use ftp://ftp.cygnus.com/pub/g++ or http://www.cygnus.com/misc//wp.
The overload resolution code was on by default. The old code could still be selected with -fno-ansi-overloading, although this was not supported.

WARNING!

progressive-96q4

progressive-97r1

The overload resolution code has been rewritten to ffb conform to the latest C++ Working Paper.
Built-in operators are now considered as candidates in operator overload resolution. Function template overloading chooses the more specialized template, and properly handles base classes in type deduction and guiding declarations. For those customers having the progressive-96q4 release, this code was not on by default and, so, they can use the -fansi-overloading flag to turn on the overload resolution code. It is, starting with the 98r1 release, now on by default.
The GNU C++ driver ( g++ ) no longer links with libg++ by default; it is now functionally identical to the C++ driver.
RTTI support has been rewritten to work properly and is now on by default.
This means code that uses virtual functions will have a modest space overhead, and will also depend on the RTTI support library code in libstdc++ . This dependency is removed. Use the -fno-rtti flag to disable RTTI support.
On ELF systems, duplicate copies of symbols with initialized common linkage like template instantiations, vtables, and extern inlines will now be discarded by the GNU linker, so - frepo isn't necessary
Partial specialization of class templates is now supported.
Synthesized destructors are no longer made virtual just because the class already has virtual functions; it is only if they override a virtual destructor in a base class. The compiler will warn if this affects your code.
(void *)0 is no longer considered a null pointer constant; NULL in <stddef.h> is now defined as __null . This is a magic constant of type, (void *) , normally, or (size_t) with -ansi .
The new template <> specialization syntax is now accepted and ignored.
The name of a class is now implicitly declared in its own scope; for example, A::A refers to A .
The STL code is based on the free SGI version, which is more efficient and conformant than the older HP distribution.
Default function arguments in templates will not be evaluated (or checked for semantic validity) unless they are needed.
The -ftemplate-depth-NN flag can be used to increase the maximum recursive template instantiation depth, defaulting to 17. If you need to use this flag, the compiler will tell you.
The internal interface between RTTI-using code and the RTTI support library has changed, so code that uses dynamic_cast should be recompiled. The RTTI support library has moved from libstdc++ to libgcc, so you no longer need to link against libstdc++ for a program that doesn't use the "hosted" library.
bool is now always the same size as another built-in type. Previously, a 64-bit RISC target using a 32-bit ABI would have 32-bit pointers and a 64-bit bool . This should not affect any supported platforms.
new (nothrow) is supported.
Some warnings were ad ffb ded for violation of style guidelines.
g++ uses an implementation of templates that are minimally parsed when seen and then later expanded. This allows conformant early name binding and instantiation controls, since instantiations no longer go through the parser.

What you get:

Instantiations of class templates and methods defined in the class body are deferred until required (unless -fexternal-templates is specified).
Nested types in class templates work.
Static data member templates work.

Possible problems:

Static data member templates work.
Types and class templates used in templates must be declared first, or the compiler will assume they are not types, and fail.
Similarly, tag nested types of template type parameters with typename .
Syntax errors in templates that are never instantiated will now be diagnosed.
Synthesized methods are emitted in any translation units that need an out-of-line copy. They are no longer affected by #pragma interface or #pragma implementation .

Local classes are supported.
The warning flag, -Wsign-compare , included in -Wall , warns about dangerous comparisons of signed and unsigned values; it was previously part of -W .
The flag, -fno-weak , disables the use of weak symbols.
The typedirective, __attribute__ , is supported.
-Woverloaded-virtual now warns if a virtual function in a base class is hidden in a derived class, rather than warning about virtual functions being overloaded (even if all of the inherited signatures are overridden).
The compiler no longer emits a warning if an ellipsis is used as a function's argument list.
Exception handling support has been significantly improved, though optimization is still not supported.
Definition of nested types outside of their containing class is supported. Use the following source code, as an example.

struct A {

       struct B;

       B* bp;

};

struct A::B {

       int member;

};

Definition of nested types outside of their containing class is suppo ffb rted. Use the following source code, as an example.
Explicit instantiation of template constructors and destructors is now supported. For example: template A<int>::A(const A&);
All HPPA targets support the -mspace option, which is experimental code aimed at reducing the size of a program at the expense of increasing execution time.
On the HPPA, some classes that do not define a copy constructor will be passed and returned in memory again so that functions returning those types can be inlined.

Debugger issues

gdb

The d10v-elf configuration includes commands to collect and display trace data.
Use 'trace' to enable tracing, 'untrace' to disable, 'info trace' to describe the information collected in the trace buffer, and 'tdisassemble' to display the list of instructions that were executed.
gdb now supports the debugging of overlays. Do 'help overlay' for more information. Example code may be found in examples/overlay.
Embedded MIPS configurations (such as mips-elf and mips-ecoff ) may use hardware breakpoints and watchpoints when communicating with boards using PMON (using the 'target pmon' command).
You may now use gdb with the Macraigor Systems "wiggler" and "OCD serial box" devices. The target commands are 'target ocd wiggler lpt1' (for PC hosts, for the wiggler device, only) and 'target ocd port' (normal serial transport; /dev/ttya ... being the pathname for addressing the variable argument for port ).
The powerpc-elf configuration includes two new target protocols; 'target sds' is an SDS-compatible protocol that is useful with Motorola's ADS821/860 boards, and 'target dink32' works with Motorola systems running the DINK32 ROM monitor (such as the embedded Yellowknife).
The 'set architecture' command allows for an explicit choice of the processor type being debugged, while 'info architecture' displays the current target architecture. Most configurations presently support only a single architecture.
The debugger was modified to properly debug executables that were compiled with the Cfront C++ version 2 compiler.
The debugger includes support for the Apple Macintosh, as a host only. GDB can be run as either an MPW tool or as a standalone application, and it can debug through the serial port. All of the usual debugger commands are available, but you must supply 'serial' as the device type to the target command, instead of '/dev/ttyXX'. Use target serial for input.
See 'mpw-README' in the main directory for more information on how to build. The MPW configuration scripts '*/mpw-config.in' support only a few targets, and only the mips-idt-ecoff target has been completely tested. Both m68k and PowerPC Macs are supported.
Use target ppcbug for support for the PowerPC ppcbug monitor.
Use target sh3 for support for the Hitachi SH3 monitor ROM.
Use the auto-solib-add variable to read in symbols from all shared libraries. If the value of auto-solib-add is 1, then symbols from all shared libraries will be read in when the program starts up. This is convenient if you want to reference a symbol in a shared library without having to stop in that library first, such as in setting a breakpoint. The default value is 0 , which improves startup time.

NOTE:

Use the command, dont-repeat, in user-defined commands to defeat the auto-repeat of gdb when an empty command is entered.
The symbol reader for AIX gdb now uses partial symbol tables. This can greatly improve startup time, especially for large executables.
When printing the type of a variable declared with a typedef, gdb uses the typedef name if possible instead of the typedef definition.
Performance is improved in MIPS IDT debugging (MIPS targets), both for stepping and for downloads.
The remotedelay option is set by default to 1. Loading executables can be considerably slower with remotedelay set to 1, but it gets around a loading bug on certain H8/300 boards.
To see a noticeable speed-up in loading when you're not using an H8/300 board, set remotedelay to 0. Use set remotedelay 0 as input.
A common hurdle in cross development is to get the communications set up properly between the target board and the development platform.
The debugger's 'set remotedebug' command can help. It was designed to help develop new remote targets; it displays the packets transmitted back and forth between the debugger and the target environment. This command can be helpful in diagnosing communications problems, for example, allowing you to observe packets not getting through or picking up noise on the line.
The set remotedebug command is now consistent among the MIPS remote target; remote targets using the gdb-specific protocol; UDI (the AMD debug protocol for the 29k ffb ); the 88k bug monitor; and Hitachi ROM monitors. You can set it to an integer specifying a protocol-debug level (normally 0 or 1 , but 2 means more protocol information for the MIPS target). See GDB and remote MIPS boards in Debugging with GDB in GNUPro Debugging Tools for details.
If you use the gcc option, '-gstabs+', gcc embeds extended debugging information in coff object files. The extended debug information is based on the stabs debugging format, which was originally used only with the a.out object file format; see The stabs debug format , in your sources as 'src/gdb/doc/stabs.texinfo', or contact Cygnus for more information. With this additional debugging information, you can debug C++ programs with gdb, even on systems that use COFF. You can get better C++ debugging by compiling with '-gstabs+' for the a29k-amd-udi, h8300-hms, m68k-coff, m88k-coff, sh-hms, and z8k-coff targets.

Assembler issues

gas

The assembler now has an -alm option, which can be used to list macro expansions. The -al c option can be used to skip false conditionals in listings.
The default objdump disassembly format has changed. You can get the old format by using --prefix-addresses .
The assembler now supports macros without requiring the assembler's preprocessor, gasp.
gas now supports the -M or --mri option, permitting the assembly of MRI-format assembler files.
The SunOS assembler is now able to assemble PIC.

Linker issues

ld

The linker now accepts the --no-whole-archive flag, to force it to not include the entire contents of an archive file.
The -rpath-link option has been added for SunOS and ELF systems.
The COFF lin ffb ker now automatically combines struct , union, and enumdebugging information, so that the information only appears once in the output file. This only applies when using COFF debugging information, as opposed to stabs.
The SunOS linker is now able to create shared libraries.

Rebuilding issues

Rebuilding From Source

GNUPro Advanced Topics

When rebuilding from source on an AIX platform, on some versions, the tr utility in '/usr/ucb' has a bug. Make sure the PATH for the build will make '/bin/tr' or '/usr/bin/tr' available ahead of '/usr/ucb/tr'. This bug is fixed in AIX 4.1.4.
There is a reported problem in rebuilding GNUPro Toolkit using AIX 3.2. x native tools. (This problem does not crop up if you use gcc to rebuild the tools.) On the RS / 6000, XLC version 1.3.0.0 miscompiles 'jump.c'. XLC version 1.3.0.1 or later fixes this problem. You can obtain xlc version 1.3.0.2 by requesting PTF 421749 from IBM. This is not relevant for AIX 4.1. 4.
Use --with-gnu-as when configuring MIPS, if you rebuild the entire GNUPro Toolkit from source.
Top-level configuration files handle the configuration for you automatically. But if you rebuild the compiler alone for a MIPS target, we highly recommend that you specify '--with-gnu-as' on the command line for configure. This avoids an incompatibility between the assembler, as, and the MIPS assembler. The MIPS assembler does not support debugging directives, and gcc uses a special program, mips-tfile, to generate them. as parses the debugging directives directly without mips-tfile .
You should also specify '--with-stabs' on the command line to configure. This provides better debugging symbols, in particular for C++. If you plan to use the linker, be sure to specify '--with-gnu-ld' when you rebuild on any platform for which the linker is available.

WARNING!

/usr/ucb/cc

you should not use

/opt/SUNWspro/bin/cc

As in previous releases, you can reconfigure GNUPro Toolkit to support more than one object format. For detailed instructions, see Rebuilding From Source in GNUPro Advanced Topics.
To add support for more object file formats (besides the format appropriate for the configured target), list the additional targets as arguments to the configure option, --enable-targets , separated by commas.
Use the following input as an example.

./configure --enable-targets=m68k-coff,i386-elf,decstation

bfd/config.bfd

--enable-targets=all

Starting with the 97r1 release, make had been altered to support the Win32 host environment better. It now has two major modes of operation:
The default mode is to support native Win32 makefiles. In this mode, Makefile rules may use either backslashes or forward slashes as filename directory separators. Makefile rules may invoke "del", "copy", or other Win32 shell builtins since make will now invoke a native Win32 subshell when necessary. Note that some programs like del do not support filenames containing forward slash directory separators so you probably want to use backslashes if you intend to use those types of programs. make does not automatically convert forward slashes to backslashes because doing so would break the use of the forward slash as the option specifier for some Win32 programs.

cmd.exe

command.com

del

cmd.exe

command.com

An optional Unix compatibility mode to support Unix-style Makefiles.
There are two ways to tell make to operate in this mode: you can either set the environment variable,MAKE_MODE, to "unix" before running make or you can invoke make as "make --unix" (in this mode, backslashes retain the usual Unix semantics and cannot be used in filenames as directory separators; however, Win32 paths without backslashes are still supported). Use of Bourne shell built-ins is permitted but Win32 shell built-ins like del and copy are unavailable. You must provide a working 274 Bourne shell for this mode to work correctly. It should be named "sh.exe" and needs to be in your path so make can find it.
After rebuilding from source, it is now necessary to rerun install-sid in order to reset your customer-id in send-pr. Previously, the default value was the value in the send-pr program in your path at the time of configuration.