Release notes for Purify 5.3 IRIX6
Contents
========
o Changes from previous releases
o Supported systems
o Restrictions and known issues
New in This Release
===================
- Bug fixes and compatibility with OS
patches.
- 64-bit applications are now supported.
A 64-bit program is one that is built
with the compiler option "-64."
- Support for Irix 7.3 compiler
- Support for gcc 2.8.1 and 2.9.5 compiler
Please read the Restrictions for important
related IRIX patch information.
New in Purify 5.2
=================
- Bug fixes and compatibility with OS
patches.
- This release uses a new FlexLm based
licensing. Read the new installation
guide before installing the product.
Use rs_install instead of pure_install
for the installation.
New in Purify 4.6
=================
- IRIX 5.x is no longer supported.
New in Purify 4.5
=================
- Support for Irix 6.5.x and the 7.2.1 compilers
- Support for applications using POSIX threads
on Irix 6.5.x. Pthread support is not
available for versions of Irix prior to 6.5.
Please read the Restrictions for important
related IRIX patch information.
- More aggressive checking of optimized N32 code
has been implemented. This can lead to more
UMR's. See the option -ignore-n32-o-umr=yes in
the Restrictions section for information on
how to disable this behavior.
- This is the last release to support Irix 5.x
==================================================
Supported systems
=================
Operating system and Hardware
-----------------------------
Purify has been tested with IRIX
versions 6.2 and 6.5.
Purify also supports 64-bit programs on
IRIX64 6.5. A 64-bit program is one that
is built with the compiler option "-64."
Compilers
---------
Purify has been tested with the
following compilers:
- cc
- CC
- f77
- gcc
- g++
Debuggers
---------
- dbx, version 6.1 or better
- cvd, the Workshop Debugger, version 2.5 or
better.
See the "Restrictions and Known Issues"
section for more details.
Threads
-------
Purify supports these threads packages:
- the sproc interface.
- POSIX threads.
==================================================
Restrictions and Known Issues
=============================
General
-------
- Purify supports both 32-bit and 64-bit
application development, and will select
the correct mode of operation automatically
based on the executable or shared object.
- The product home directory has been
reorganized to support both 32 and 64-bit
development. This organization should be
transparent for all 32-bit users and most
64-bit users. However, the location of
the Purify stubs library is different.
32-bit libraries have been moved to the
lib32 directory:
purifyhome/lib32/libpurify_stubs.a
purifyhome/lib32/libpurify_stubs.so
purifyhome/lib32/libpurify_stubs_n32.a
purifyhome/lib32/libpurify_stubs_n32.so
To preserve backward compatibility, the
following links are provided in purifyhome:
purifyhome/libpurify_stubs.a
purifyhome/libpurify_stubs.so
purifyhome/libpurify_stubs_n32.a
purifyhome/libpurify_stubs_n32.so
32-bit API users are encouraged to use
the libraries from the lib32 directory,
and not from purifyhome.
64-bit API users must link against the
equivalent library in the lib64 directory:
purifyhome/lib64/libpurify_stubs.a
purifyhome/lib64/libpurify_stubs.so
The API header file has not moved and is
shared for both development modes.
- If you use Purify -printhomedir to
prefix product libraries, you must qualify
the library path with the appropriate lib
sub-directory:
`purify -printhomedir`/lib32/libpurify_stubs.a
`purify -printhomedir`/lib32/libpurify_stubs.so
`purify -printhomedir`/lib32/libpurify_stubs_n32.a
`purify -printhomedir`/lib32/libpurify_stubs_n32.so
`purify -printhomedir`/lib64/libpurify_stubs.a
`purify -printhomedir`/lib64/libpurify_stubs.so
- You may not see Purify error messages from an
MPI (Message Passing Interface) test program.
An MPI program first creates a given number of
children and these children actually execute
the test program. The parent doesn't execute
the test program.
So, one needs to set the purify option
-follow-child-processes=yes to get the errors
from an MPI test program. Note that when you
set this option, a new purify viewer per child
will come up. Thus, the number of processes
increases. This could have an effect on the
performance of your program. So, it is better
to test with smaller number of children.
User Interface
--------------
- In version 5.3, the format of view files has
changed. This means that the 5.3 Viewer cannot
be used to view older view files, and 5.3 view
files cannot be opened by older Viewers.
Please use the Viewer of the same version
to open view files.
- If a large number of items are selected,
"Expand all" followed by "Collapse all" can
crash some unpatched versions of the
OpenWindows 3.0 server.
This occurs if you are displaying on a SUN
workstation.
- If you expand or collapse messages while the
"Continue" or "Reset etc. Continue" buttons
are displayed, the buttons may subsequently be
incorrectly positioned.
- The "Edit" toolbar item may be slow to
respond.
- The Purify GUI menus and buttons become
inaccessible if either the NumLock or
ScrollLock key is activated. The workaround
is to switch them off, or add the following
line(s) to your $HOME/.Xdefaults file.
! Ignore the NumLock and ScrollLock keys on
! mouse buttons
Purify*ignoreModifierMask: Mod3|Mod2
This second workaround will take effect for
a new Purify viewer after you restart
your X-session or run a command like
'xrdb -merge $HOME/.Xdefaults'.
- The "Invoke ClearDDTS" Button has been
modified to bring up the ClearQuest web
interface. This feature only works with
Netscape Navigator.
The site-wide URL for ClearQuest can be
given during installation or set by
manually editing the file
pure_clearquest_url
in your Purify home directory. A user
can override the site-wide URL by setting
the environment variable
PURE_CLEARQUEST_URL
This feature is partly implemented by a shell
script, ("pure_invoke_clearquest" in your
Purify home directory) to allow you to
tailor its operation to your needs. If you
wish, you may copy and customize this.
script. As long as the directory containing
the script appears in your search path
before your Purify home directory, it will
be used instead of the original script.
If you prefer to use Purify with
ClearDDTS, you can do so by setting the
X resource:
Purify*ddtsCommandString
to 'xddts', if xddts is in your search path,
or to the full path to your xddts executable.
xddts is invoked by a shell script
("pure_invoke_ddts" in your Purify home
directory). If you wish to customize it,
please read the section on customizing
"pure_invoke_clearquest" above.
If you already have a customized
"pure_invoke_ddts" script in your search path,
All you need to do is set your X resource as
described above, and Purify will find
your customized script automatically.
The following copyright applies to portions
of this ClearQuest integration code:
Copyright 1996 Netscape Communications
Corporation, all rights reserved. Created:
Jamie Zawinski , 24-Dec-94.
Permission to use, copy, modify, distribute,
and sell this software and its documentation
for any purpose is hereby granted without
fee, provided that the above copyright
notice appear in all copies and that both
that copyright notice and this permission
notice appear in supporting documentation.
No representations are made about the
suitability of this software for any
purpose. It is provided "as is" without
express or implied warranty.
Compilers
---------
- Those purified programs compiled with gcc
which initialize function pointers globally
may dump core. g++ programs with global
constructors may also dump core.
This is due to a bug (#801456) in SGI
assembler. SGI will make it available in
Irix 6.5.10 and will also release a
patch (#3944) for the previous Irix
releases. Please contact SGI Support to
obtain this patch.
Note that the g++ library libstdc++ may
have been built by the old assembler.
Please build this with the new assembler
to ensure that the purified program runs
without dumping.
- The GNU gcc extensions are not tested against
Purify. Most gcc extensions will
probably work fine. Known limitations at
present include problems with nested functions
(e.g.: making a pointer to a nested function
and attempting to call through it will not
work).
- For systems running IRIX 6.5.7f or higher,
Purify may fail to instrument /usr/lib/libc.so
with the error message
Processing /usr/lib/libc.so.1 - error fatal:
Jump registered out of no transformed block
[SA906D8, SA90710] /usr/lib/libc.so.1
This is a bug in cord. libc.so is corded,
but cord is not updating the notransform
information in the shared library correctly.
This does not happen with Irix versions
below 6.5.7, or with version on the
maintenance stream (e.g. 6.5.7m).
Rational Support is working with SGI to make
a patch available. Please contact Rational
Support regarding availability of the patch.
- Aggressive Loading
- Some compilers load data from memory but
ignore the data that has been read,
especially when optimization is turned on.
Purify will signal a UMR if the loaded
data is uninitialized. In some sense this
is a false error report because the
uninitialized data will not affect your
program.
Purify'ing X Applications
-------------------------
- When running a Purify'd X application, there
is a potential for deadlock if your
application causes Purify to generate a
message while the application is holding the X
lock, since Purify will be unable to generate
the message, and the application is blocked
until the message is delivered.
To avoid this kind of problem, you should run
your application on a different X server than
the Purify UI or Purify stderr output, or you
should use the -log-file= or -view-file=
options to specify a file to capture messages
for inspection after your application is
finished.
A convenient way to debug on two displays is
to pre-start the Purify Viewer on one display
("slave"), and then start the application on
the other display ("master"):
% purify -display slave:0 -view a.out.X &
% a.out.X -display master:0
The two commands must be executed on the same
computer, but it could be the workstation
associated with either display, or altogether
another computer remote from both displays.
The application will connect to the already
started Purify Viewer, and messages will not
conflict with the X display interactions of
the application under test.
Debuggers
---------
- Invoking a debugger on a core file
generated by a purified program, will
cause numerous "mismatch" messages
to be printed. These can be ignored,
and will not affect debugging.
Threads
-------
- Call chains describing when memory was
malloced or freed do not always include the
thread id.
- The Purify API functions purify_map_pool() and
purify_map_pool_id() are not MT safe.
- For applications using Pthreads, the purified
program may crash or hang if dlopen is called.
This is a bug in rld. SGI released
patch (#3910) to correct the problem; please
contact SGI Support to obtain the patch.
Note that your application may not use
dlopen directly, but a library it is using
may call dlopen. C++ exception handling may
also call dlopen.
- Customers using unsupported threads packages
should contact Rational Software technical
support (support@rational.com) to ensure
compatibility.
Unsupported Features
--------------------
- SBR and SBW errors are not reported on IRIX.
- The Initialized Data and BSS sections are not
checked for ABR and ABW errors on IRIX.
- Statically linked mallocs are ignored: users
who implement their own malloc function must
put it in a DSO.
- Instrumentation of stripped binaries and
shared libraries is not supported.
Misleading Error Messages
-------------------------
- A failure to write to disk may result in a
message like:
#Error: (system) No such file or directory,
write() failed [a.out].
In this case, you can see that you are out of
disk space with "df .", and you should clean
up or otherwise acquire more disk space.
- More aggressive UMR checking has been added.
This may cause spurious UMR's in some
optimized n32 code. To disable this aggressive
checking, use the option:
-ignore-n32-o-umr=yes
or set the environment variable:
PURIFY_IGNORE_N32_O_UMR
Purify Swap Usage
-----------------
A Purify'd application needs more swap space
than the equivalent unpurify'd version. Running
under a debugger, you will need even more. To
add swap space you may do the following:
Do the following as root to add a 40MB swap
file:
/usr/sbin/mkfile 40m /usr/swap
Add this line to /etc/fstab:
/usr/swap swap swap pri=3 0 0
Then execute:
/etc/init.d/swap stop
/etc/init.d/swap start
You may also allocate virtual swap to appease
the debuggers swap requirements. Check the man
page (man swapctl) for caveats.
Do the following as root:
vi /etc/config/vswap.options
>change vswaplen to 250000
/sbin/chkconfig vswap on
/etc/init.d/swap stop
/etc/init.d/swap start
How do I use the API functions?
------------------------------
There are stubs for the Purify API functions in
the files libpurify_stubs.so and
libpurify_stubs.a. For the N32 ABI, the files
are libpurify_stubs_n32.so and
libpurify_stubs_n32.a. You use the .so during
development, and the .a for shipping your
product. If you link with the .a file, the
Purify API functions will be stubbed out even if
you subsequently Purify your program.
Manual Examples
---------------
The examples in Chapter 12 of the manual are
intended to help the user understand the
circumstances that would cause Purify to report
what it reports. At least one example does not
apply to Purify on IRIX.
- The MSE example in Chapter 12 will not produce
an MSE on IRIX.