NetBSD/gnu/dist/gcc/FAQ

589 lines
27 KiB
Plaintext

GCC Frequently Asked Questions
The latest version of this document is always available at
[1]http://gcc.gnu.org/faq.html.
This FAQ tries to answer specific questions concerning GCC. For
general information regarding C, C++, resp. Fortran please check the
[2]comp.lang.c FAQ, [3]comp.std.c++ FAQ, and the [4]Fortran
Information page.
Other GCC-related FAQs: [5]libstdc++-v3, and [6]GCJ.
_________________________________________________________________
Questions
1. [7]General information
1. [8]What is the relationship between GCC and EGCS?
2. [9]What is an open development model?
3. [10]How do I get a bug fixed or a feature added?
4. [11]Does GCC work on my platform?
2. [12]Installation
1. [13]How to install multiple versions of GCC
2. [14]Dynamic linker is unable to find GCC libraries
3. [15]libstdc++/libio tests fail badly with --enable-shared
4. [16]GCC can not find GNU as/GNU ld
5. [17]cpp: Usage:... Error
6. [18]Optimizing the compiler itself
7. [19]Why does libiconv get linked into jc1 on Solaris?
3. [20]Testsuite problems
1. [21]How do I pass flags like -fnew-abi to the testsuite?
2. [22]How can I run the test suite with multiple options?
4. [23]Older versions of GCC
1. [24]Is there a stringstream / sstream for GCC 2.95.2?
5. [25]Miscellaneous
1. [26]Friend Templates
2. [27]dynamic_cast, throw, typeid don't work with shared
libraries
3. [28]Why do I need autoconf, bison, xgettext, automake, etc?
4. [29]Why can't I build a shared library?
5. [30]When building C++, the linker says my constructors,
destructors or virtual tables are undefined, but I defined
them
6. [31]Will GCC someday include an incremental linker?
_________________________________________________________________
General information
What is the relationship between GCC and EGCS?
In 1990/1991 gcc version 1 had reached a point of stability. For the
targets it could support, it worked well. It had limitations inherent
in its design that would be difficult to resolve, so a major effort
was made to resolve those limitations and gcc version 2 was the
result.
When we had gcc2 in a useful state, development efforts on gcc1
stopped and we all concentrated on making gcc2 better than gcc1 could
ever be. This is the kind of step forward we wanted to make with the
EGCS project when it was formed in 1997.
In April 1999 the Free Software Foundation officially halted
development on the gcc2 compiler and appointed the EGCS project as the
official GCC maintainers. The net result was a single project which
carries forward GCC development under the ultimate control of the
[32]GCC Steering Committee.
_________________________________________________________________
What is an open development model?
We are using a bazaar style [33][1] approach to GCC development: we
make snapshots publicly available to anyone who wants to try them; we
welcome anyone to join the development mailing list. All of the
discussions on the development mailing list are available via the web.
We're going to be making releases with a much higher frequency than
they have been made in the past.
In addition to weekly snapshots of the GCC development sources, we
have the sources readable from a CVS server by anyone. Furthermore we
are using remote CVS to allow remote maintainers write access to the
sources.
There have been many potential GCC developers who were not able to
participate in GCC development in the past. We want these people to
help in any way they can; we ultimately want GCC to be the best
compiler in the world.
A compiler is a complicated piece of software, there will still be
strong central maintainers who will reject patches, who will demand
documentation of implementations, and who will keep the level of
quality as high as it is today. Code that could use wider testing may
be integrated--code that is simply ill-conceived won't be.
GCC is not the first piece of software to use this open development
process; FreeBSD, the Emacs lisp repository, and the Linux kernel are
a few examples of the bazaar style of development.
With GCC, we are adding new features and optimizations at a rate that
has not been done since the creation of gcc2; these additions
inevitably have a temporarily destabilizing effect. With the help of
developers working together with this bazaar style development, the
resulting stability and quality levels will be better than we've had
before.
[1] We've been discussing different development models a lot over
the past few months. The paper which started all of this introduced
two terms: A cathedral development model versus a bazaar
development model. The paper is written by Eric S. Raymond, it is
called ``The Cathedral and the Bazaar''. The paper is a useful
starting point for discussions.
_________________________________________________________________
How do I get a bug fixed or a feature added?
There are lots of ways to get something fixed. The list below may be
incomplete, but it covers many of the common cases. These are listed
roughly in order of decreasing difficulty for the average GCC user,
meaning someone who is not skilled in the internals of GCC, and where
difficulty is measured in terms of the time required to fix the bug.
No alternative is better than any other; each has its benefits and
disadvantages.
* Fix it yourself. This alternative will probably bring results, if
you work hard enough, but will probably take a lot of time, and,
depending on the quality of your work and the perceived benefits
of your changes, your code may or may not ever make it into an
official release of GCC.
* [34]Report the problem to the GCC bug tracking system and hope
that someone will be kind enough to fix it for you. While this is
certainly possible, and often happens, there is no guarantee that
it will. You should not expect the same response from this method
that you would see from a commercial support organization since
the people who read GCC bug reports, if they choose to help you,
will be volunteering their time.
* Hire someone to fix it for you. There are various companies and
individuals providing support for GCC. This alternative costs
money, but is relatively likely to get results.
_________________________________________________________________
Does GCC work on my platform?
The host/target specific installation notes for GCC include
information about known problems with installing or using GCC on
particular platforms. These are included in the sources for a release
in INSTALL/specific.html, and the [35]latest version is always
available at the GCC web site. Reports of [36]successful builds for
several versions of GCC are also available at the web site.
_________________________________________________________________
Installation
How to install multiple versions of GCC
It may be desirable to install multiple versions of the compiler on
the same system. This can be done by using different prefix paths at
configure time and a few symlinks.
Basically, configure the two compilers with different --prefix
options, then build and install each compiler. Assume you want "gcc"
to be the latest compiler and available in /usr/local/bin; also assume
that you want "gcc2" to be the older gcc2 compiler and also available
in /usr/local/bin.
The easiest way to do this is to configure the new GCC with
--prefix=/usr/local/gcc and the older gcc2 with
--prefix=/usr/local/gcc2. Build and install both compilers. Then make
a symlink from /usr/local/bin/gcc to /usr/local/gcc/bin/gcc and from
/usr/local/bin/gcc2 to /usr/local/gcc2/bin/gcc. Create similar links
for the "g++", "c++" and "g77" compiler drivers.
An alternative to using symlinks is to configure with a
--program-transform-name option. This option specifies a sed command
to process installed program names with. Using it you can, for
instance, have all the new GCC programs installed as "new-gcc" and the
like. You will still have to specify different --prefix options for
new GCC and old GCC, because it is only the executable program names
that are transformed. The difference is that you (as administrator) do
not have to set up symlinks, but must specify additional directories
in your (as a user) PATH. A complication with --program-transform-name
is that the sed command invariably contains characters significant to
the shell, and these have to be escaped correctly, also it is not
possible to use "^" or "$" in the command. Here is the option to
prefix "new-" to the new GCC installed programs:
--program-transform-name='s,\\\\(.*\\\\),new-\\\\1,'
With the above --prefix option, that will install the new GCC programs
into /usr/local/gcc/bin with names prefixed by "new-". You can use
--program-transform-name if you have multiple versions of GCC, and
wish to be sure about which version you are invoking.
If you use --prefix, GCC may have difficulty locating a GNU assembler
or linker on your system, [37]GCC can not find GNU as/GNU ld explains
how to deal with this.
Another option that may be easier is to use the --program-prefix= or
--program-suffix= options to configure. So if you're installing GCC
2.95.2 and don't want to disturb the current version of GCC in
/usr/local/bin/, you could do
configure --program-suffix=-2.95.2 <other configure options>
This should result in GCC being installed as /usr/local/bin/gcc-2.95.2
instead of /usr/local/bin/gcc.
_________________________________________________________________
Dynamic linker is unable to find GCC libraries
This problem manifests itself by programs not finding shared libraries
they depend on when the programs are started. Note this problem often
manifests itself with failures in the libio/libstdc++ tests after
configuring with --enable-shared and building GCC.
GCC does not specify a runpath so that the dynamic linker can find
dynamic libraries at runtime.
The short explanation is that if you always pass a -R option to the
linker, then your programs become dependent on directories which may
be NFS mounted, and programs may hang unnecessarily when an NFS server
goes down.
The problem is not programs that do require the directories; those
programs are going to hang no matter what you do. The problem is
programs that do not require the directories.
SunOS effectively always passed a -R option for every -L option; this
was a bad idea, and so it was removed for Solaris. We should not
recreate it.
However, if you feel you really need such an option to be passed
automatically to the linker, you may add it to the GCC specs file.
This file can be found in the same directory that contains cc1 (run
gcc -print-prog-name=cc1 to find it). You may add linker flags such as
-R or -rpath, depending on platform and linker, to the *link or *lib
specs.
Another alternative is to install a wrapper script around gcc, g++ or
ld that adds the appropriate directory to the environment variable
LD_RUN_PATH or equivalent (again, it's platform-dependent).
Yet another option, that works on a few platforms, is to hard-code the
full pathname of the library into its soname. This can only be
accomplished by modifying the appropriate .ml file within
libstdc++/config (and also libg++/config, if you are building libg++),
so that $(libdir)/ appears just before the library name in -soname or
-h options.
_________________________________________________________________
GCC can not find GNU as/GNU ld
GCC searches the PATH for an assembler and a loader, but it only does
so after searching a directory list hard-coded in the GCC executables.
Since, on most platforms, the hard-coded list includes directories in
which the system assembler and loader can be found, you may have to
take one of the following actions to arrange that GCC uses the GNU
versions of those programs.
To ensure that GCC finds the GNU assembler (the GNU loader), which are
required by [38]some configurations, you should configure these with
the same --prefix option as you used for GCC. Then build & install GNU
as (GNU ld) and proceed with building GCC.
Another alternative is to create links to GNU as and ld in any of the
directories printed by the command `gcc -print-search-dirs | grep
'^programs:''. The link to `ld' should be named `real-ld' if `ld'
already exists. If such links do not exist while you're compiling GCC,
you may have to create them in the build directories too, within the
gcc directory and in all the gcc/stage* subdirectories.
GCC 2.95 allows you to specify the full pathname of the assembler and
the linker to use. The configure flags are `--with-as=/path/to/as' and
`--with-ld=/path/to/ld'. GCC will try to use these pathnames before
looking for `as' or `(real-)ld' in the standard search dirs. If, at
configure-time, the specified programs are found to be GNU utilities,
`--with-gnu-as' and `--with-gnu-ld' need not be used; these flags will
be auto-detected. One drawback of this option is that it won't allow
you to override the search path for assembler and linker with
command-line options -B/path/ if the specified filenames exist.
_________________________________________________________________
cpp: Usage:... Error
If you get an error like this when building GCC (particularly when
building __mulsi3), then you likely have a problem with your
environment variables.
cpp: Usage: /usr/lib/gcc-lib/i586-unknown-linux-gnulibc1/2.7.2.3/cpp
[switches] input output
First look for an explicit '.' in either LIBRARY_PATH or
GCC_EXEC_PREFIX from your environment. If you do not find an explicit
'.', look for an empty pathname in those variables. Note that ':' at
either the start or end of these variables is an implicit '.' and will
cause problems.
Also note '::' in these paths will also cause similar problems.
_________________________________________________________________
Optimizing the compiler itself
If you want to test a particular optimization option, it's useful to
try bootstrapping the compiler with that option turned on. For
example, to test the -fssa option, you could bootstrap like this:
make BOOT_CFLAGS="-O2 -fssa" bootstrap
_________________________________________________________________
Why does libiconv get linked into jc1 on Solaris?
The Java front end requires iconv. If the compiler used to bootstrap
GCC finds libiconv (because the GNU version of libiconv has been
installed in the same prefix as the bootstrap compiler), but the newly
built GCC does not find the library (because it will be installed with
a different prefix), then a link-time error will occur when building
jc1. This problem does not show up so often on platforms that have
libiconv in a default location (like /usr/lib) because then both
compilers can find a library named libiconv, even though it is a
different library.
Using --disable-nls at configure-time does not prevent this problem
because jc1 uses iconv even in that case. Solutions include
temporarily removing the GNU libiconv, copying it to a default
location such as /usr/lib/, and using --enable-languages at
configure-time to disable Java.
_________________________________________________________________
Testsuite problems
How do I pass flags like -fnew-abi to the testsuite?
If you invoke runtest directly, you can use the --tool_opts option,
e.g:
runtest --tool_opts "-fnew-abi -fno-honor-std" <other options>
Or, if you use make check you can use the make variable RUNTESTFLAGS,
e.g:
make RUNTESTFLAGS="--tool_opts '-fnew-abi -fno-honor-std'" check-g++
_________________________________________________________________
How can I run the test suite with multiple options?
If you invoke runtest directly, you can use the --target_board option,
e.g:
runtest --target_board "unix{-fPIC,-fpic,}" <other options>
Or, if you use make check you can use the make variable RUNTESTFLAGS,
e.g:
make RUNTESTFLAGS="--target_board 'unix{-fPIC,-fpic,}'" check-gcc
Either of these examples will run the tests three times. Once with
-fPIC, once with -fpic, and once with no additional flags.
This technique is particularly useful on multilibbed targets.
_________________________________________________________________
Older versions of GCC and EGCS
Is there a stringstream / sstream for GCC 2.95.2?
Yes, it's at:
[39]http://gcc.gnu.org/ml/libstdc++/2000-q2/msg00700/sstream.
_________________________________________________________________
Miscellaneous
Friend Templates
In order to make a specialization of a template function a friend of a
(possibly template) class, you must explicitly state that the friend
function is a template, by appending angle brackets to its name, and
this template function must have been declared already. Here's an
example:
template <typename T> class foo {
friend void bar(foo<T>);
}
The above declaration declares a non-template function named bar, so
it must be explicitly defined for each specialization of foo. A
template definition of bar won't do, because it is unrelated with the
non-template declaration above. So you'd have to end up writing:
void bar(foo<int>) { /* ... */ }
void bar(foo<void>) { /* ... */ }
If you meant bar to be a template function, you should have
forward-declared it as follows. Note that, since the template function
declaration refers to the template class, the template class must be
forward-declared too:
template <typename T>
class foo;
template <typename T>
void bar(foo<T>);
template <typename T>
class foo {
friend void bar<>(foo<T>);
};
template <typename T>
void bar(foo<T>) { /* ... */ }
In this case, the template argument list could be left empty, because
it can be implicitly deduced from the function arguments, but the
angle brackets must be present, otherwise the declaration will be
taken as a non-template function. Furthermore, in some cases, you may
have to explicitly specify the template arguments, to remove
ambiguity.
An error in the last public comment draft of the ANSI/ISO C++ Standard
and the fact that previous releases of GCC would accept such friend
declarations as template declarations has led people to believe that
the forward declaration was not necessary, but, according to the final
version of the Standard, it is.
_________________________________________________________________
dynamic_cast, throw, typeid don't work with shared libraries
The new C++ ABI in the GCC 3.0 series uses address comparisons, rather
than string compares, to determine type equality. This leads to better
performance. Like other objects that have to be present in the final
executable, these std::typeinfo_t objects have what is called vague
linkage because they are not tightly bound to any one particular
translation unit (object file). The compiler has to emit them in any
translation unit that requires their presence, and then rely on the
linking and loading process to make sure that only one of them is
active in the final executable. With static linking all of these
symbols are resolved at link time, but with dynamic linking, further
resolution occurs at load time. You have to ensure that objects within
a shared library are resolved against objects in the executable and
other shared libraries.
* For a program which is linked against a shared library, no
additional precautions need taking.
* You cannot create a shared library with the "-Bsymbolic" option,
as that prevents the resolution described above.
* If you use dlopen to explicitly load code from a shared library,
you must do several things. First, export global symbols from the
executable by linking it with the "-E" flag (you will have to
specify this as "-Wl,-E" if you are invoking the linker in the
usual manner from the compiler driver, g++). You must also make
the external symbols in the loaded library available for
subsequent libraries by providing the RTLD_GLOBAL flag to dlopen.
The symbol resolution can be immediate or lazy.
Template instantiations are another, user visible, case of objects
with vague linkage, which needs similar resolution. If you do not take
the above precautions, you may discover that a template instantiation
with the same argument list, but instantiated in multiple translation
units, has several addresses, depending in which translation unit the
address is taken. (This is not an exhaustive list of the kind of
objects which have vague linkage and are expected to be resolved
during linking & loading.)
If you are worried about different objects with the same name
colliding during the linking or loading process, then you should use
namespaces to disambiguate them. Giving distinct objects with global
linkage the same name is a violation of the One Definition Rule (ODR)
[basic.def.odr].
For more details about the way that GCC implements these and other C++
features, please read the [40]ABI specification. Note the
std::typeinfo_t objects which must be resolved all begin with "_ZTS".
Refer to ld's documentation for a description of the "-E" &
"-Bsymbolic" flags.
_________________________________________________________________
Why do I need autoconf, bison, xgettext, automake, etc?
If you're using diffs up dated from one snapshot to the next, or if
you're using the CVS repository, you may need several additional
programs to build GCC.
These include, but are not necessarily limited to autoconf, automake,
bison, and xgettext.
This is necessary because neither diff nor cvs keep timestamps
correct. This causes problems for generated files as "make" may think
those generated files are out of date and try to regenerate them.
An easy way to work around this problem is to use the gcc_update
script in the contrib subdirectory of GCC, which handles this
transparently without requiring installation of any additional tools.
(Note: Up to and including GCC 2.95 this script was called egcs_update
.)
When building from diffs or CVS or if you modified some sources, you
may also need to obtain development versions of some GNU tools, as the
production versions do not necessarily handle all features needed to
rebuild GCC.
In general, the current versions of these tools from
[41]ftp://ftp.gnu.org/gnu/ will work. At present, Autoconf 2.50 is not
supported, and you will need to use Autoconf 2.13; work is in progress
to fix this problem. Also look at
[42]ftp://gcc.gnu.org/pub/gcc/infrastructure/ for any special versions
of packages.
_________________________________________________________________
Why can't I build a shared library?
When building a shared library you may get an error message from the
linker like `assert pure-text failed:' or `DP relative code in file'.
This kind of error occurs when you've failed to provide proper flags
to gcc when linking the shared library.
You can get this error even if all the .o files for the shared library
were compiled with the proper PIC option. When building a shared
library, gcc will compile additional code to be included in the
library. That additional code must also be compiled with the proper
PIC option.
Adding the proper PIC option (-fpic or -fPIC) to the link line which
creates the shared library will fix this problem on targets that
support PIC in this manner. For example:
gcc -c -fPIC myfile.c
gcc -shared -o libmyfile.so -fPIC myfile.o
_________________________________________________________________
When building C++, the linker says my constructors, destructors or virtual
tables are undefined, but I defined them
The ISO C++ Standard specifies that all virtual methods of a class
that are not pure-virtual must be defined, but does not require any
diagnostic for violations of this rule [class.virtual]/8. Based on
this assumption, GCC will only emit the implicitly defined
constructors, the assignment operator, the destructor and the virtual
table of a class in the translation unit that defines its first such
non-inline method.
Therefore, if you fail to define this particular method, the linker
may complain about the lack of definitions for apparently unrelated
symbols. Unfortunately, in order to improve this error message, it
might be necessary to change the linker, and this can't always be
done.
The solution is to ensure that all virtual methods that are not pure
are defined. Note that a destructor must be defined even if it is
declared pure-virtual [class.dtor]/7.
_________________________________________________________________
Will GCC someday include an incremental linker?
Incremental linking is part of the linker, not the compiler. As such,
GCC doesn't have anything to do with incremental linking. Depending on
what platform you use, it may be possible to tell GCC to use the
platform's native linker (e.g., Solaris' ild(1)).
References
1. http://gcc.gnu.org/faq.html
2. http://www.eskimo.com/~scs/C-faq/top.html
3. http://www.jamesd.demon.co.uk/csc/faq.html
4. http://www.fortran.com/fortran/info.html
5. http://gcc.gnu.org/onlinedocs/libstdc++/faq/index.html
6. http://gcc.gnu.org/java/faq.html
7. http://gcc.gnu.org/faq.html#general
8. http://gcc.gnu.org/faq.html#gcc
9. http://gcc.gnu.org/faq.html#open-development
10. http://gcc.gnu.org/faq.html#support
11. http://gcc.gnu.org/faq.html#platforms
12. http://gcc.gnu.org/faq.html#installation
13. http://gcc.gnu.org/faq.html#multiple
14. http://gcc.gnu.org/faq.html#rpath
15. http://gcc.gnu.org/faq.html#rpath
16. http://gcc.gnu.org/faq.html#gas
17. http://gcc.gnu.org/faq.html#environ
18. http://gcc.gnu.org/faq.html#optimizing
19. http://gcc.gnu.org/faq.html#iconv
20. http://gcc.gnu.org/faq.html#testsuite
21. http://gcc.gnu.org/faq.html#testoptions
22. http://gcc.gnu.org/faq.html#multipletests
23. http://gcc.gnu.org/faq.html#old
24. http://gcc.gnu.org/faq.html#2.95sstream
25. http://gcc.gnu.org/faq.html#misc
26. http://gcc.gnu.org/faq.html#friend
27. http://gcc.gnu.org/faq.html#dso
28. http://gcc.gnu.org/faq.html#generated_files
29. http://gcc.gnu.org/faq.html#picflag-needed
30. http://gcc.gnu.org/faq.html#vtables
31. http://gcc.gnu.org/faq.html#incremental
32. http://gcc.gnu.org/steering.html
33. http://gcc.gnu.org/faq.html#cathedral-vs-bazaar
34. http://gcc.gnu.org/bugs.html
35. http://gcc.gnu.org/install/specific.html
36. http://gcc.gnu.org/buildstat.html
37. http://gcc.gnu.org/faq.html#gas
38. http://gcc.gnu.org/install/specific.html
39. http://gcc.gnu.org/ml/libstdc++/2000-q2/msg00700/sstream
40. http://www.codesourcery.com/cxx-abi/
41. ftp://ftp.gnu.org/gnu/
42. ftp://gcc.gnu.org/pub/gcc/infrastructure/