Sun Studio FAQ

h2. Introduction

So far, this FAQ list is mostly composed of things I culled from the Sun Studio
forums, but I'm hoping it will become more than that with the help of
you and other Sun Studio users.
If you don't see your question answered here, you can ask it in on the Sun Studio forums at :
http://developers.sun.com/sunstudio/community/index.jsp, or you can add it
here without an answer, and someone will probably fill in the answer for you.
Don't forget you can add pages to your "watchlist" to get email updates
when a page changes.

Remember, the best and most complete
information about Sun Studio can always be found on the developer portal
at http://developers.sun.com/sunstudio.

There has been some discussion of having a wiki site of some kind available
on sun.com. If that happens, this content might eventually move to that
new site. Stay tuned.

{toc}

h1. Installer

h2. Installer - How to install as non-root (on Linux)

The installer can not run as a user other than root. However, you can use
rpm2cpio on Linux to extract the contents of the RPMs and place them in a
directory of your choice.

Some releases of Sun Studio are available as tarballs. Those can be
untarred wherever you want.

h2. Installer - how to uninstall

As root:
{noformat}
% cd /var/sadm/prod/com.sun.studio_11
% ./batch_uninstall_all
{noformat}


h2. Installer - Complaint about libXp.so on 64-bit linux

You might need to install the 32-bit Motif libraries by hand.
See here : [Installing Sun Studio on Different Linux Distros]

h2. Installer - no bash?

If you install Solaris without SUNWbash, then the "sunstudio" script won't be
able to start the IDE. More importantly, the installer won't let you
install the Sun Studio tools unless that package is available. The installer
may just hang if this is missing.

The only way I can think of this happening is if you install Solaris 8
with the "minimal" option. Don't do that.

You can use this command to check for bash.

{noformat}
% pkgchk -l -p /usr/bin/bash
Pathname: /usr/bin/bash
Type: regular file
Expected mode: 0555
Expected owner: root
Expected group: bin
Expected file size (bytes): 425948
Expected sum(1) of contents: 49550
Expected last modification: Jan 05 02:21:03 PM 2000
Referenced by the following packages:
SUNWbash
Current status: installed
{noformat}

If you are missing this package, you should be able to
find it on CD 2 of the base Solaris 8 installation CD's.
I'm not sure, but the original S8 FCS version might not have
had it, so it might require a somewhat recent update of Solaris 8.

There is also bad information floating around that says to install
bash from the Companion CD. This is wrong. That's a different
package and won't satisfy the package dependency required by the
installer.

h2. Installer - can't get -nodisplay to work

I've seen this problem reported multiple times, but I'm not
sure there is just one answer. Here is one possible answer:

http://forum.sun.com/jive/thread.jspa?forumID=309&threadID=96020

From Terry, our installer guru.
{noformat}
It should be

# ./installer -nodisplay

Make sure to set your $DISPLAY to a valid value. it will still run in
command-line mode with the right flag. Another possibility is that
you have an older version of java first on your path. Try typing
java -version as root, and verify that you are picking up the
version you think you are (it has to be greater than 1.4.2 for
the nodisplay stuff to work properly). Finally, if none of
these suggestions help, please run

sh -x ./installer -nodisplay

and post the output, so we can debug where the shellscript is making the mistake.

--Terry.
{noformat}

h2. Installer - How to install on Ubuntu

Thanks to Matt: http://forum.sun.com/jive/thread.jspa?forumID=318&threadID=102562

{noformat}

If you are still interested in installing Sun Studio 11 under
Ubuntu Dapper Drake you can try installing the rpm package
using apt-get then making a symbolic link to the command
in the /bin directory. In other words run

apt-get install rpm
ln -s /usr/bin/rpm /bin/rpm
After this the regular installer worked for me.

I am not sure if its important but I also have gawk and the java jre and jdk installed.

Regards,
Matt

{noformat}

See also: [Installing Sun Studio on Different Linux Distros]

h2. Installer - I pkgrm'ed everything, but I still can't reinstall.

From akumta on the developer forums:

Removing the packages manually (via pkgrm) has probably
resulted in corrupting the productregistry file.

Please take a look at the Product Registry Problems
section in the installation guide at
http://docs.sun.com/source/816-7874/troubleshooting.html

The original thread is here:
http://forum.sun.com/jive/thread.jspa?forumID=309&threadID=102761

h2. Installer - 0 bytes will be installed

If you try to install and see lots of information like this:

{noformat}
Product: Sun Studio Software
Location: /u01
Size: 0 bytes
----------------------------
{noformat}

Then perhaps you have gotten bits for the wrong architecture.
Trying to install the SPARC product on an x86 machine can
result in this kind of problem. This is considered a bug,
since the problem should be diagnosed much sooner.

h1. General Issues

h2. General - What are the latest patches?

The list of patches for Sun Studio 11 is here:
http://developers.sun.com/prodtech/cc/downloads/patches/ss11_patches.html

The sun.com web site can sometimes take a while to get updated
with the latest list of available patches.

See also [Sun Studio Patches]

h2. General - Why does updatemanager fail to install Sun Studio patches?

updatemanager can fail with this error
{noformat}
Utility used to install the update failed with exit code {0}.
{noformat}

It's a problem related to global zones and using patchadd -G.

You can find out more here:

http://forum.java.sun.com/thread.jspa?threadID=5107410

Until this gets worked out you have install patches
without the help of updatemanager. I think this problem
can be fixed with better metadata in the patches, so when
the Sun Studio patches all get revved again they should
all start to work with updatemanager.

It's very likely that the latest patches for the patch tools
will help to fix this problem. Can anyone verify this?
{noformat}
New Solaris 10 patchadd patches resolve a few of the compiler patch installation issues:

6188748 patchadd needs support for patching from NFS filesystems
6374972 patchadd should deal with global-only packages intelligently

Especially, if you install Sun Studio 11 or 12 on a Solaris 10 system
which has the global zone only, you no longer need to use "patchadd -G"
to get the patches installed.

There is a problem with 119254-33/119255-33, please use at least the -34 revisions of the patchadd patches.
(Thanks Thomas Dehn)
{noformat}

h2. General - How do I report a bug in Sun Studio

{noformat}
Visit bugs.sun.com
Select "Report a Bug" under Bug Database list on the right hand side of the page.
Check the box under header "Submit a Bug", and click on button "Start a new Report"
Under *Product/Category *pulldown menu, you will see 8 entries for eight components
under the main one "Sun Studio Compilers and Tools":
C Compiler
C++ Compiler
Fortran
Dbx
Performance Analyzer
IDE
Dmake
Other

You'll recieve an acknowledgment of your submital, and if accepted
and entered into Sun's bug database an email with a Bug ID to track
the issue with.
{noformat}

h2. General - How do I search existing bugs in Sun Studio

This feature is coming soon, we're working on it.

h2. General - Does Sun Studio work with the latest and greatest Solaris?

The Sun Studio tools go through some pretty serious testing before we release
the final FCS versions, but Sun can only test against platforms that are
available when the testing is done. In general Sun tries to fix bugs that crop
up when you're using newer versions of Solaris and Java.

Here is a list of which versions of Sun Studio were officially tested with
which versions of Solaris.

http://developers.sun.com/prodtech/cc/support/support_matrix.jsp

See also: [Sun Studio Solaris Issues]

h2. General - Does Sun Studio work with the latest and greatest Java?

(See related question about Solaris, above)

See also: [Sun Studio Java Issues]

h2. General - Why are the tools sending packets or emails outside my site!

There are features in Sun Studio to try and tell Sun how many people
are using Sun Studio. There are two mechanisms at work, and both can
be disabled.

One older mechanism is based on sending emails, and is only enabled
in pre-release versions of the tools. You can disable that one by setting
an environement variable called UT_NO_USAGE_TRACKING with a value of "1".
The newer one is based on sending a simple packet every two weeks or so,
and is turned on for pre-release tools and for officially supported releases.
The new one can be disabled by setting SUNW_NO_UPDATE_NOTIFY to any value
other than "false".

If sendmail is not set up correctly on your system, it's likely that
you'll have problems with dbx. It may hang when it tries to send mail.
If this happens, just set the environment variable listed above.

The newer packet based system is used to find out how many users we have,
and also to give you information about
the latest updates. You can read more about it here:
http://developers.sun.com/prodtech/cc/documentation/ss11/release_notes.html#Update

It's very useful for Sun to know which of the individual tools are being used more
often, and which platforms are being used. It helps Sun decide which features to
implement next, so please leave them on if you can. It will encourage us to
support and enhance the tools that are important to you.

h1. Tools - IDE

h2. IDE - How do I add new file suffixes as "source files"

In Sun Studio 11 there is no mechanism for augmenting the list
of file extensions treated as "source files". In Sun Studio 12
you can use the GUI to add new extensions by following this path:

Tools->Options->Advanced Options->IDE Configuration->System->Object Types

h2. IDE - increasing the font size

From Nik M

Try "sunstudio --fontsize 18".
It will change the font size for all windows except Editor.
Editor fonts can be changed inside the IDE:
- main menu: Tools->Options->"Fonts & Colors" tab.

h2. IDE - The Sun Studio IDE is ignoring my control keys

{noformat}
Ivan:
We've had two queries now on the development tools forum where
users are complaining that control keys like BS, DEL, arrows are ignored
by the SunStudio IDE. This apparently happens for all Swing widgets.

Anyone recognize this issue?
{noformat}

{noformat}
This smells like they are using an X server which has the
XKEYBOARD extension enabled. Run xdpyinfo to check for that.
There also have been a few locale specific issues like this.
Its a java issue, you can confirm that by using the
notepad demo application that ships with J2SE.

# cd /<java_path>/demo/jfc/Notepad
# java -jar Notepad.jar

Fixed in Mustang, its a whole bushel of bugs.


With kind regards,


Thomas Dehn
{noformat}

{noformat}
Editor's note: This may be related to copy/paste problems too:
IvanIgorovich:
There is a bug in Java, 6173972, that may be affecting this.
It's fixed in Java5.0u3, so java version SS is running under
and see if an upgrade will help.
{noformat}

You might also want to try this simple workaround if
you encounter this bug.

{noformat}
From: Phil R.
my reading of it is that it is specific to the XAWT toolkit.
Whilst this is now the default on Solaris and Linux, selecting
the older MAWT toolkit is a simple matter of specifying a system
property or an environment variable at JDK start up time.
eg
setenv AWT_TOOLKIT sun.awt.motif.MToolkit
{noformat}

h2. IDE - I get an error about [: too many arguments

This happens in Sun Studio 11 if the output of uname -p has multiple words in it.
It should be fixed in newer versions of the sunstudio startup script.
You can specify the user directory explicitly as a workaround.

--userdir /some/other/dir

http://forum.java.sun.com/thread.jspa?threadID=5105057

h2. IDE - How do I update or convert my Sun Studio 'project' files?

http://forum.java.sun.com/thread.jspa?forumID=852&threadID=5181160

Jens Siebert has written a Python script for parsing project files, that
might be helpful.

http://home.arcor.de/jsiebert/files/ss12-gtk-project.py

Check the above site for any updates, but I'll copy the script to

h2. Dbx - binding arrow keys

* how do I bind the arrow keys in the dbx command window?

{noformat}

# black magic to set arrow keys bindings
bind ^[[=prefix-1
bind ^[[A=up-history
bind ^[[B=down-history
bind ^[[D=backward-char
bind ^[[C=forward-char
# clear screen on Ctrl-L
bind -m ^L='/usr/bin/clear^J'
# if you have trouble with backspace/delete
bind ^?=delete-char-backward

# Thanks also to Nikolay Igotti and Fui Shien for
# helping this information find its way to the dbx group who
# should have known this in the first place. ;-)

# Thanks to Volker Simonis for tip about backspace


{noformat}

h2. Dbx - "Mismatched checksum ..."

Why does dbx issue "Mismatched checksum ..." messages when loading system libraries?

Warnings are generated during .gnu_debuglink section processing, which contains
a pointer to a separate file with debug information and checksum of this file. If
the checksum does not match the one that dbx calculates, dbx does not process this
external debug information and issues warning like this:

dbx: internal warning: Mismatched checksum - file /lib/tls/i686/cmov/libc-2.3.6.so
has cd229517, expecting c99d524a

This does not prevent dbx neither from debugging nor from seeing exported
library symbols (such as global variables and functions). Extended
debugging information, however, will not be available.

These warnings have been seen on 32-bit Ubuntu Dapper Drake LTS (6.06)
and Ubuntu Feisty Fawn (7.04), Nexenta (GNU/OpenSolaris).

NOTE: this appeared to be due to differences between dbx and binutils (2.17)
in finding external debug info algorithm. Dbx code has been adjusted to match
the one in binutils, fix should be available in next but one patch for dbx.

See also: "Separate symbol files" forum thread here* *http://forum.java.sun.com/thread.jspa?threadID=5072470&tstart=30

h2. Dbx - annoying arrow at the end of the line

What is that arrow at the end of the line? Can't dbx edit multi-line commands?

{noformat}
Is there any way to change the printable length of command line on dbx?
When we types a long command, it could be automatically wrapped.

For example, when I type the following long command,

(dbx) /usr/ccs/bin/elfdump /usr/openwin/lib/locale/ja/xlibi18n_ja.so.2 | grep PLTGOT

the first of command is wrappd and actually displayed like follows.

(dbx) /locale/ja/xlibi18n_ja.so.2 | grep PLTGOT <

I know I can use *history* command as a workaround to see the whole command line..
But it's more convenient for me if I can set the printable length... Any idea..?
{noformat}

The dbx ksh is showing its age, and wrapping around the current line being edited is one
of the features it doesn't have. In the long run we want to update or replace the
built-in ksh with something better. You can use these commands:

{noformat}
(dbx) set +o vi
(dbx) set +o emacs
{noformat}

Which will turn off cmdline editing and allow wrapping.

h2. Dbx - working with Pro*C

We've had several forum users asking questions about how to get Pro*C to work with
Sun Studio.

http://forum.sun.com/jive/thread.jspa?forumID=309&threadID=95976

http://forum.sun.com/jive/thread.jspa?threadID=72482&tstart=30

http://forum.sun.com/jive/thread.jspa?threadID=72483&tstart=30

http://forum.sun.com/jive/thread.jspa?threadID=72478&tstart=30

The last link above has a brief description of a common problem
with missing source and line number information in dbx and
the debugger GUI.

{noformat}

Use the command line option LINES=YES to proc
to get #line directives in the generated C or C++ code.

See the Pro*C/C++ Precompiler Programmer's Guide available online
(http://www.oracle.com/technology/documentation/index.html)

{noformat}

h2. Dbx - my stack trace is messed up on x86

You have to use the -xregs=no%frameptr option with the Sun Studio compilers
if you want to be able to get a stack trace. (The -xregs option tells
the optimizer which registers it can use for program variables. So to
preserve the ABI value of the frame pointer, you tell the optimizer not
to use it for program variables.)

On x64 (aka amd64) there
is an ABI-defined data structure (called the .eh_frame section) that
supports walking the stack even without
storing the frame pointer. Some tools like dbx and the C++ runtime will use
this ABI information, but some tools won't.

On 32-bit x86, there is no alternate way to walk the stack. The code has to
be compiled with a frame pointer in order to do a stack trace. The only way
this optimization will get turned on for you is when the -fast option is
used to enable the most aggressive level of optiization. If you want the
aggressive optimization, and stack traces, then use -fast -xregs=no%frameptr.

You can read more about this option (and some related documentation bugs) here.

http://blogs.sun.com/roller/page/dew?entry=the_real_story_on_the

h2. Dbx - The JVM keeps getting a SIGILL

The JVM implementation from Sun uses several signals internally to itself as part
of its design. SIGILL is one of those signals. Normally dbx intercepts signals
that it thinks users will want to see. In this case you should tell dbx to ignore
SIGILL signals when you are debugging the JVM. Use the command "ignore sigill"
from the dbx command line.

h2. Dbx - RTC keeps giving me RUA errors in string routines on Solaris 10

There is a dbx patch to fix this problem.

http://blogs.sun.com/dbx/entry/latest_updates_to_run_time

dbx patch 121023-02 -- Bug 6372403

h2. Dbx - Why does my program run differently under dbx?

Dbx has several side-effects on your program when you run it.

[Dbx Side Effects]

h2. Dbx - Why can't I redirect the output of some dbx commands?

The desired meaning of ">" is different in these two commands:

(dbx) print (j > 5)

(dbx) print foo > /tmp/output

Dbx just isn't smart enough to figure out which one you meant, so
some commands interpret ">" as a C/C++ operator, and some commands
assume that ">" is a redirection operator. You can read more
about it in the dbx command line help system. Type "help redirection".

h1. Compilers

h2. OpenMP - Why can't my loops have MT-safe functions?

The optimizer doesn't know how you're going to link your program.
Just because you called a function called "malloc" doesn't mean
that's going to be an MT-safe malloc when you link your program.
You can tell the compiler to assume you're not replacing library
functions by using the -xbuiltin flag. (See the manual)

For calls to user functions, cross-file analysis is needed to
figure out if your function is MT-safe. You can try -xO5 -xipo
(or -xipo=2). Those options might help.

h2. Compilers - Why do I see : "language optional software package not installed"

Don't put /usr/ucb first in your search path. Modify the PATH
environment variable so that /usr/ucb comes last, or remove that
part of your PATH variable. The file where this configuration
information is stored depends on what your login shell is. Try looking
in a file named ".cshrc" in your home directory.

h2. Compilers - When I use dwarf my executable is now huge!

See this page : [Dwarf size increase]

h2. Compilers - Why does the same source produce different object files?

There are some options that cause local symbols to be made
global. When the compiler does that it has to add some kind of
hash prefix so that the symbols don't clash with each other.

See this page : [Compiler Globalization]

There is also a date stamp that gets recorded in the debug
information, when you compile with -g. stripping the binary
files will remove that, but won't remove affects of globalization.

See also : http://sunsolve.sun.com/search/document.do?assetkey=1-25-80040

h2. C++ - I'm having trouble with the Boost libraries

See here : [Boost and Sun Studio]

h2. C++ - Why is my symbol missing?

Sometimes a C++ program will fail to link because of a missing symbol, but the
symbol looks like something that should already be defined. Steve Clamage provided
a good summary of what this can mean.

{noformat}
The most common cause for a missing function is that a needed binary is not being
linked, or that the definition is #ifdef'd away.

Another possibility is that function cleanupConfig is not declared consistently
everywhere. Some examples:

1. The declaration for the function in a header looks like
int cleanupConfig(_config*);
but in the definition looks like
int cleanupConfig(const _config*) { ... }

2. The function is declared inline in some places and not inline in other
places. I've seen code like this:
INLINE int foo(int i) { ... }
where INLINE was a macro that expanded to "inline" or to nothing. If INLINE is
not defined the same way for every declaration of this function foo, you get
inconsistent definitions, which can lead to link failures.

3. The parameter to cleanupConfig is declared via a typedef:
typedef _config cleanup_kind;
int cleanupConfig(cleanup_kind*);
and the definition of cleanup_kind is not identical everywhere.

4. The function is declared at global scope, but the function is defined inside a namespace.
{noformat}

h2. C++ - I have more questions!

There are more Sun Studio C++ FAQ's (relevant to Sun Studio 11) here:

http://developers.sun.com/sunstudio/documentation/ss11/mr/READMEs/c++_faq.html

h2. C - What preprocessor symbols are predefined in C ?

(From Douglas:)

------------------------------------------------

Use the -### (or -xdryrun) options to see a detailed list of the -D options
passed to the C compiler front-end. These are the predefined
macros. The only other macros predefined that are not passed
to the C compiler front-end are those required by the ISO C standard:

{noformat}
__STDC__ Linux: always expands to 1
......................Solaris: expands to 1 with -Xc
.....................................expands to 0 with -Xa, -Xt
.....................................not defined with -Xs

__STDC_VERSION__ Expands to 199901L with -xc99=all
.....................Expands to 199409L with -xc99=none
.....................Not defined with -Xs option

__STDC_IEC_559__ Expands to 1
__STDC_IEC_559_COMPLEX__ Expands to 1
__STDC_HOSTED__ Expands to 1
{noformat}
------------------------------------------------

h1. Platform Specific

h2. Linux - I get a strange complaint about a shared library

{noformat}
% cc -g -xdebugformat=dwarf r.c
"r.c", line 4: internal compiler error: DBGGEN ERROR: FILE="../src/dbg_libdwarf.c", LINE=46, Could not load dwarf library:
libdwarf.so : libdwarf.so: cannot open shared object file: No such file or directory (DBG_GEN 5.1.11)
cc: acomp failed for r.c
{noformat}

You might need a patch for your linux system to fix the behavior of the dlopen() call.

http://blogs.sun.com/roller/page/quenelle?entry=linux_compilers_require_a_glibc

h2. Linux - The linker keeps saying: "mangled line number section"

{noformat}
I've been getting errors like this:

pcsLP.o(.text+0x4167)/usr/bin/ld: Dwarf Error: mangled line number section.
: In function `unsigned long PcsPtrace::get_rtld_base_addr(char*)':
: warning: the use of `tmpnam' is dangerous, better use `mkstemp'

This happens when the linker tries to look up the line/file location
of a relocation or other link-time error. You only see the complaint
about line number information when the linker is also trying to report
another error.

It looks like a bug in gnu ld that was fixed on March 6th.

The fix is shown in revision 1.86 of file dwarf2.c in the 'binutils' package.
http://sourceware.org/cgi-bin/cvsweb.cgi/src/bfd/dwarf2.c?cvsroot=src

I assume this symptom will go away in some future version of gnu ld.
{noformat}

In some cases the Sun Studio 12 compilers on Linux will use a private
copy of the GNU linker to link your program. The version of ld
included with Sun Studio was chosen to match our supported platforms,
and it's a little bit old. You can try telling the Sun Studio compilers
to use your local version of the linker if you have a recent version.

For C++, you can use: "CC -Yl,/usr/bin" to use the system linker instead.
Using the system linker is the normal case, I think only C++ uses the
private linker, in order to make sure we have one that supports COMDAT
section groups (which we use for template instances).

This message comes from the linker because the linker is trying to use Dwarf information
to translate a link-time error address into a source file name. It's very
clever and useful when it works, but if it encounters something that it
thinks is an error in the dwarf data, it spits out this warning.

h2. Linux - problem linking with libm.so
C++ had a problem with default linker on RH4 not being able to grok our object files due to its own bugs.
Thats why we bring updated linker with us (.../prod/lib/amd64/ld or something like this) and use it in 64bit compilation.
C does not have this problem and thus does not use this linker.

Some linuxes (like RH5 or SuSE10) have recently changed its object files format which causes our "older" linker to emit this "file not recognized" message. Typical error message is
{noformat}
$ CC -m64 hello.c
/usr/lib64/libm.so: file not recognized: File format not recognized
{noformat}
but it is not limited to libm.so.

Source code of linker patched by Sun can be found here:
http://javashoplm.sun.com/ECom/docs/Welcome.jsp?StoreId=8&PartDetailId=SS12-SRCID-MAY07-G-F&TransactionId=Try

See these forum thread for more information:

http://forum.java.sun.com/thread.jspa?forumID=850&threadID=5181629

http://forum.java.sun.com/thread.jspa?forumID=855&threadID=5178078

h2. Linux - sunstudio -- too many arguments

{noformat}
sunstudio: line 293: [: too many arguments -- Gentoo AMD64
Error: client VM not supported
{noformat}

This happens if the output of "uname -p" results in a string that has spaces in it.
If you supply the --userdir option to sunstudio it will override the
the usage of uname, and it should work okay.

http://forum.sun.com/jive/thread.jspa?forumID=318&threadID=105057

h2. Linux - sunstudio -- On x64 Linux, IDE fails with UnsatisfiedLinkError

{noformat}
java.lang.UnsatisfiedLinkError: /opt/sun/sunstudio11/prod/lib/libbase.so: /opt/sun/sunstudio11/prod/lib/libbase.so:
... <more stack trace> ...
{noformat}

If the Sun Studio 11 IDE tries to start up using a 64-bit JVM, it will
fail to start because it won't be able to load some of the native
support libraries that are 32-bit. A bug has been filed for this (6475414).

There are some steps for downloading and using a 32-bit JVM
supplied by geo.mac on this thread:

http://forum.sun.com/jive/thread.jspa?threadID=106103&tstart=0