|
HP C++ Version 7.3
HP C++ Version 7.3
Release Notes for OpenVMS Alpha
June 14, 2007
This document contains information about new and changed features in
this version of HP C++ for OpenVMS Alpha.
Hewlett-Packard Company Palo Alto, California
© Copyright 2007 Hewlett-Packard Development Company, L.P.
Confidential computer software. Valid license from HP required for
possession, use or copying. Consistent with FAR 12.211 and 12.212,
Commercial Computer Software, Computer Software Documentation, and
Technical Data for Commercial Items are licensed to the U.S. Government
under vendor's standard commercial license.
The information contained herein is subject to change without notice.
The only warranties for HP products and services are set forth in the
express warranty statements accompanying such products and services.
Nothing herein should be construed as constituting an additional
warranty. HP shall not be liable for technical or editorial errors or
omissions contained herein.
UNIX is a registered trademark of The Open Group.
Portions of the ANSI C++ Standard Library have been implemented using
source licensed from and copyrighted by Rogue Wave Software, Inc. All
rights reserved.
Information pertaining to the C++ Standard Library has been edited and
reprinted with permission of Rogue Wave Software, Inc. All rights
reserved.
Portions copyright 1994-2005 Rogue Wave Software, Inc.
1 Introduction
This document contains the release notes for HP C++ Version 7.3
for OpenVMS Alpha. The HP C++ product requires OpenVMS Alpha Version
7.3-2 or higher.
The HP C++ media contains the following:
- The Version 7.3 kit, which includes the HP C++ compiler and
associated files, such as the Class Library and Standard Library header
files.
HTML files are provided for the release notes and some of the product
manuals for use with a web browser. To view this documentation, point
your browser to
file:/sys$common/syshlp/cxx$help/index.htm
|
2 Important Compatibility Information
HP strives to maintain a high degree of compatibility between
successive versions of the compiler and its run-time environment.
Because, however, each new version includes enhancements and changes,
you should be aware of the following whenever you upgrade:
- Differences between Standard Library versions
- Differences between compiler versions
- Differences between HP C++ and the C++ International Standard
- Retirement of
cfront
language dialect
The next sections discuss these differences.
2.1 Standard Library Differences
Any code that references the C++ Standard Library (any of the STL
containers or algorithms, standard iostreams or locales) that was
compiled using version V6.2 or earlier of the HP C++ compiler must be
recompiled and relinked in order to be used with the code compiled with
HP C++ Version 6.3 or later.
For example, if you have an object library, an object module, or a
shareable image compiled with HP C++ Version 6.2, you must to recompile
it (and relink in the case of a shareable image) before an application
compiled with Version 6.3 or later can use it.
As stated earlier, this restriction applies only to code that
references the C++ Standard Library; it does not apply to code that
references the pre-standard Class Library, because the stability of
that library's interface guarantees link compatibility in future
releases.
Beginning with HP C++ Version 6.3, the Standard Library is guaranteed
to be link-compatible in all subsequent releases.
For applications that will relink on the end-user's system, see
Deploying Your Application
in
HP C++ User's Guide for OpenVMS Systems
for information about redistributing C++ Run-Time Library components.
2.2 Compiler Differences
Starting with Version 6.0, the HP C++ compiler differs significantly from previous versions. There are several major differences that you should be aware of before using a Version 6.n or higher compiler for the first time. These differences a
Appendix E Compiler Compatibility
in
HP C++ User's Guide for OpenVMS Systems
.
- Language differences
The compiler implements (with some
differences, as described in Section 2.3), the C++ International
Standard, which differs significantly from the language specified in
the ARM (The Annotated C++ Reference Manual, 1991, by Ellis and Stroustrup) and implemented
by the Version 5.n compilers. When switching from a Version
5.n compiler, you might need to modify your source files,
especially if you use the default language mode. In addition, language
changes can affect the run-time behavior of your programs. If you want
to compile existing source code with minimal source changes, compile
using the
/standard=arm
option. See
Appendix E Compiler Compatibility
in
HP C++ User's Guide for OpenVMS Systems
.
Note
The installation procedure checks whether a Version 5.3 to 5.6 compiler
is installed on your system. If so, it asks whether you want to save
it, and if so, where. The default save area for a Version 5.6 compiler
is SYS$COMMON:[CXX056].
If you find that Version 6.n or higher requires excessive
changes to your applications even when you use the
/standard=arm
option, or if you encounter problems using the Version 6.n or
higher compiler, you can return to your previous C++ environment by
invoking the command procedure
SYS$COMMON:[CXX0nn]COMPILER_SETUP.COM
|
where nn specifies the version of your previous compiler.
|
- Diagnostic differences
The Version 6.n or higher
compiler does more error checking than Version 5.6 and generates more
diagnostics. If you want the number of diagnostics issued by the
Version 6.n or higher compiler to be similar to Version 5.6,
compile with the
/quiet
option. See
Message Control and Information Options
in
HP C++ User's Guide for OpenVMS Systems
.
- Implementation differences
The automatic template instantiation model has been redesigned for the current version. Although code compiled with a Version 5.n compiler and a Version 6.n or higher compiler can be combined,
Using Templates
section in the Compiler Compatibility Appendix of the
HP C++ User's Guide for OpenVMS Systems
.
2.3 Differences Between HP C++ and the C++ International Standard
The following items, specified in the C++ International Standard, are
not supported in Version 7.3 but may be supported in a future version.
Details about each item are provided in the indicated sections of the
C++ International Standard document and The C++ Programming Language, 3rd Edition by Bjarne
Stroustrup.
- The
export
keyword for templates (Standard §14, paragraph 6;
Stroustrup §9.2.3)
2.4 Retirement of CFRONT Language Dialect
The CFRONT dialect was provided for migrating code from the CFRONT
compilers to the HP C++ compiler. Because it has been over five years
since the last CFRONT compiler was released, we are retiring this
dialect. It is removed from Version 7.1 of the compiler.
3 C++ Standard Library
This Standard Library string class, known as the String
Library, is not the same as the String Package, which
is part of the Class Library implemented in earlier versions of
HP C++.
For information about the HP C++ Class Library, See
Appendix D
in in
HP C++ User's Guide for OpenVMS Systems
.
Thread Safety
The Standard Library provided with this release is thread safe but not
thread reentrant. Thread safe means that all library internal and
global data is protected from simultaneous access by multiple threads.
In this way, internal buffers as well as global data like
cin
and
cout
are protected during each individual library operation. Users, however,
are responsible for protecting their own objects.
According to the C++ standard, results of recursive initialization are
undefined. To guarantee thread safety, the compiler inserts code to
implement a spinlock if another thread is initializing local static
data. If recursive initialization occurs, the code deadlocks even if
threads are not used.
4 Release Notes for the V7.3 C++ Compiler
The following are enhancements, changes, and restrictions for the C++
compiler environment in this version:
- A problem has fixed which could sometimes cause the compiler to
crash when using the _MB builtin functions with /OPT=LEVEL=2 and higher.
- The compiler previously asserted when processing a function that
belonged in a overload set, but which had not yet resolved to the
specific function selected by the overload, and its parameters were not
pointers or references. This has been fixed.
- A recent fix exposed another bug where the compiler failed to
appropriately mark a pointer-to-member function call which returns a
class with a non-trivial copy constructor by value, so the 'this'
pointer and the pointer to the temporary created for the class object,
were passed in an inverted order. This too has been fixed.
- The compiler bug where pointers of the correct size were not
created in 64-bit modes, has been fixed. This bug typically manifested
in the ?: conditional operator expressions, when either the second or
the third operand of the expression dereferenced a pointer expression.
- When creating a pointer-to-member call, and the pointed member
function was a function that returned a value via a copy constructor,
the synthesized call would generate an ACCVIO when run, because the
compiler did not correctly handle the transformed argument (for the
value returned by the copy constructor). This has been fixed.
- Two other problems where the compiler was not generating pointers
of the correct size in 64-bit modes have been fixed.
- Known Problem: In some situations, the compiler's loop unrolling
optimization might generate memory accesses in the code stream that
never actually execute at run-time, but which would violate the ANSI
Aliasing rules if they were to execute. In such a situation, the
compiler might emit an erroneous BADANSIALIASn message, where n is a
number or is omitted.
If the violations take place only in machine
instructions that will not execute at run-time, these messages can be
safely ignored. To determine whether or not particular instances of
a BADANSIALIASn message are erroneous, recompile the module with the
/OPT=UNROLL=1 qualifier. Any BADANSIALIASn messages that disappear
under that qualifier can be safely ignored, so you may want to add
appropriate "#pragma message" directives to the source, localized to
the specific source lines known to be safe. This is preferable to
disabling the message for the whole compilation, since in all other
cases the message indicates a real potential for code generation that
will not work as intended. And this is generally preferable to
disabling the ANSI_ALIAS or loop unrolling optimizations, since that
would likely degrade performance, although the amount of degradation is
not predictable, and in unusual cases it might even improve
performance. As always, when making changes to performance-critical
code, it's best to measure the impact.
- In previous versions of the compiler, specifying /STAN=STRICT on
the command-line disabled the /NOWARNINGS qualifier. This is no longer
done.
- The compiler now emits a warning when encountering __declspec
specifiers, which were never supported, but were silently ignored.
- The compiler incorrectly failed to compile code constructs with
__builtin_va_start when in the /STANDARD=GNU mode. This has been fixed.
- In 64-bit compilations only, the compiler would produce incorrect
code if a pointer to a const object was deferenced, its constness cast
away with a const_cast to a non-const reference of the same object
type, and an attempt was made to bind a value to this reference object.
This has been fixed.
- The compiler now diagnoses some cases where long pointers were
converted to short pointers, where previously these were not diagnosed.
- The compiler no longer emits incorrect MAYLOSEDATA diagnostics for
some simple expressions, such as assigning the address of a local
variable to a short pointer.
- Previously, the compiler incorrectly created short pointers when
arrays were decayed in certain expressions. This has been fixed.
- The compiler now implements the partial ordering specification
resolved by Core Issue 214 of the C++ Standard.
This is
demonstrated by the following program:
extern "C" int printf(const char *, ...);
typedef unsigned int size_t;
template <typename T>
void func1(const T&) { printf("T\n"); }
template <typename charT, size_t N>
void func1(charT(&)[N]) { printf("charT[N]\n"); }
template <typename X, typename T>
void func3(const T&) { printf("T\n"); }
int main() {
func1("hello");
func3<int>("hello"); // error before
}
|
- The compiler previously generated incorrect source location
information, which made the debugger behave as if the user said
"step/into" when using the command "step". This made stepping over
inlined constructors and thunks particularly difficult. This has been
fixed.
- When using the /MODEL=ANSI object model, the compiler sometimes
passed to member functions the "this" pointer and the "pointer to
struct" returned by value, in an incorrect order. This has been fixed.
- The compiler incorrectly produced an OPNDNOTCLS diagnostic for
certain code constructs when compiling in the /STANDARD=GNU dialect.
This has been fixed.
enum ia64_instruction_constants {
instruction0_shift = 5
};
class uint128_t {
unsigned lo;
public:
uint128_t(unsigned l) : lo(l) { }
unsigned low() { return lo; }
operator unsigned() const { return lo; }
uint128_t operator <<(int sa);
};
void insert_break() {
uint128_t break_raw(167);
(break_raw <<
instruction0_shift).low();
}
|
- The compiler in the /MODEL=ANSI object model, sometimes created the
"this" pointer of the incorrect size for base classes that are
templatized. This has been fixed.
- When a predefined macro is undefined using an #undef directive, the
compiler now issues the NOUNDEFPREMAC diagnostic with Warning severity
in all modes except /STANDARD=STRICT. Previously, this diagnostic had
Error severity.
- The compiler now issues a NEVERDEF diagnostic with the severity of
a discretionary error when compiling with /STANDARD=STRICT, and with
Warning severity in all other /STANDARD dialects:
struct S { inline void foo(); void bar(); }
void S::bar() { foo(); } // bar is not called, but foo is "used"
|
- The compiler now issues the BLTINCLNK diagnostic only if the
routine would have been treated as an intrinsic function had it had "C"
linkage. Previously, the diagnostic was emitted for a broader set of
routines.
- The compiler now has a fix for a buffer overflow problem that
previously was exposed when using the /XREF qualifier.
- The compiler no longer allows specifying the /PREPROCESS_ONLY and
the /OBJECT qualifiers together on the command line. Previously, the
compiler accepted both qualifiers but did not produce an object file.
- The compiler would previously ignore the specification of the
/ASSUME=NOHEADER_TYPE command-line qualifier when using the
CXX$USER_INCLUDE logical. This has been fixed.
- The compiler would sometimes produce an assertion if union objects
were passed by reference. This has been fixed.
- When compiling /OPT/DEBUG, the compiler could accvio when
processing some compiler-generated code for interludes. This has been
fixed.
- When compiling and linking /DEBUG, sometimes a static data member
was not instantiated, resulting in a undefined message from the linker.
This has been fixed.
- In rare cases, a compilation that used a comma list to specify more
than one source file could cause a compiler crash. This has been fixed.
- When /STANDARD=STRICT (or LATEST) was specified, if a type was a
dependent type, it would not accept more than one string literal
initializer in an aggregate initializer context. This has been
corrected.
template <typename T> void test() {
const T *s[] = { "foo", "bar" }; // previously an error
const char *t[] = { "foo", "bar" }; }
int main() {
const char *s[] = { "foo", "bar" };
test<char>();
}
|
- It is now possible to
#undef
a predfined macro (such as
__LINE__
) in all modes except /STANDARD=STRICT or LATEST. A warning diagnostic
is issued, but the macro will be undefined. This matches the behavior
of C++ on OpenVMS Integrity.
- The V7.1 compiler would emit a BLTINCLNK diagnostic if a C++
linkage function declared a name that would normally be an intrinsic
function, even if the paramters and/or return type did not match those
expected by the intrinsic. With this update kit, the diagnostic will
only be output if the parameter and return type matches the intrinsic.
- The V7.1 compiler would allow a command line to specify both
/PREPROCESS_ONLY and /OBJECT. In this case, the compiler would silently
ignore the /OBJECT qualifier, and no .OBJ file would be produced. With
this release, a %DCL-W-CONFLICT diagnostic is output for this case.
- With this update kit, the compiler now supports variadic macros.
This feature allows macros to take a variable number of arguments. This
feature was added to the HP C Compiler in the V6.4 release and is
supported by a number other C and C++ compilers. The feature is
available only when the value of the /STANDARD qualifier is RELAXED
(the default), MS, or GNU.
- Certain bad uses of the
va_start
macro could crash the compiler. This has been corrected.
- The PRAGIGNORE diagnostic is no longer output for the
#pragma define_template
directive when the /TEMPLATE=NOPRAGMA qualifier is specified on the
command line.
- Use of certain expressions that used the array index operator on a
string constant such as
(long)&"str"[1]
could crash the compiler. This has been corrected.
- The compiler now accepts the
__inline
language extension keyword in all language modes except
/STANDARD=STRICT (or LATEST). The
__inline__
keyword is also accepted, except in the strict or MS modes.
- There were two very minor flaws in the 64-bit pointer support. Both
involved creating 64-bit pointers using constant values. An example is:
void *n3 = (void *) 0x80000010ULL;
|
In these cases, the V7.1 compiler could sometimes generate
incorrect code when the pointers were used. This has been corrected.
- In certain cases, the compiler would emit a MAYLOSEDATA diagnostic
when it should not have. This could only happen when mixed pointer-size
support was enabled. The compiler would sometimes create a 64-bit
pointer when it should have created a 32-bit pointer. This problem has
been corrected.
- The DEFAULTTMPLARG is now a warning in all modes but the strict
modes. This matches the behavior of C++ on OpenVMS Integrity.
5 Release Notes for the V7.3 C++ Standard Library
The following are enhancements, changes, and restrictions for the C++
Standard Library in this version:
- When compiled in standard GNU mode, the library now defines the
_RWSTD_NO_IMPLICIT_INCLUSION macro which causes library headers to
#include
their respective template definition files. This is necessary because
in standard GNU mode, implicit inclusion is disabled.
Before this
change, the program below would link with undefined symbol when
compiled in standard GNU mode:
#include <vector>
int main() {
std::vector<int> v;
v.push_back(0);
}
|
6 Release Notes for the Version 7.1 C++ Compiler
The following sections describe enhancements, changes, and restrictions
for the C++ compiler environment.
6.1 Enhancements, Changes, and Problems Corrected in Version 7.1
- The /TEMPLATE_DEFINE qualifier now requires an option.
- The compiler issues a CC-W-NOTINCRTL message when it prefixes a
name not in the current C RTL.
/PREFIX_LIBRARY_ENTRIES=ALL_ENTRIES
prefixes all functions defined by the C99 standard including those that
may not be supported in the current run-time library. So calling
functions introduced in C99 that are not yet implemented in the OpenVMS
C RTL will produce unresolved references to symbols prefixed by DECC$
when the program is linked. The compiler now issues a CC-W-NOTINCRTL
message when it prefixes a name that is not in the current C RTL.
- #pragma module module-name [module-ident | "module-ident"]
If
the module-name is too long:
- A warning is generated if /NAMES=TRUNCATED is specified.
- There is no warning if /NAMES=SHORTEN is specified.
A shortened external name incorporates all the characters in the
original name. If two external names differ by as little as one
character, their shortened external names will be different. If
the optional module-ident or "module-ident" is too long, a warning is
generated. The default module-name is the filename of the first
source file. The default module-ident is "V1.0" They are treated as if
they were specified by a #pragma module directive. If the
module-name is longer than 31 characters and:
- /NAMES=TRUNCATE is specified, truncate the module-name to 31
characters, or less if the 31st character is within a Universal
Character Name.
- /NAMES=SHORTENED is specified, shorten the module-name to 31
characters using the same special encoding as other external names.
Lowercase characters in the module-name are converted to uppercase
only if /NAMES=UPPERCASE is specified. A module-ident that is
longer than 31 characters is treated as if /NAMES=(TRUNCATED,AS_IS)
were applied, truncating it to 31 characters, or less if the 31st
character is within a Universal Character Name. The default
module-name comes from the source file name which always appears in the
listing header. The module-name (and ident) appear in the listing
header only if they come from a #pragma module directive or differ from
the default.
- Certain OpenVMS conditions normally result in the delivery of
signals that can be processed using a C-style signal-handler mechanism.
The C++ condition-handling facility has been changed to invoke this
C-style signal-handling mechanism only if a signal handler has been
established. Before this change, unhandled signals would result in
program termination. In addition, the following C signals are now
recognized: SIGCHLD, SIGPIPE, SIGWINCH, and SIGSEGV. This now makes C++
for OpenVMS Alpha functionally equivalent to C++ for I64 systems.
- Specifying a C++ headers library and a C headers library using "+"
and the /LIB qualifier on the
cxx
command line, as in the following example, can cause the compiler to
fetch a C header file from the C headers library instead of a
template-definition file from the C++ headers library:
cxx x.cxx+SYS$LIBRARY:CXXL$ANSI_DEF.TLB/LIB+SYS$LIBRARY:DECC$RTLDEF.TLB/LIB
|
This can happen if a C header file has the same filename as the C++
template-definition file: for example, the
string.h
header file in the C headers library and
string.cc
template-definition file in the C++ headers library.
- The compiler was not correctly handling break statements out of
loops which follow an identifier label within switch statements. This
is now fixed.
- A bug in the C++ compiler in the /STANDARD=STRICT mode of
compilation, used to produce an error diagnostic when compiling code
that used
setjmp
or
c$$establish
. This has been fixed.
- In the /STANDARD=STRICT mode of compilation, the compiler used to
issue a diagnostic with the severity of Error, for NULL reference
expression within a
sizeof
expression. The severity of the diagnostic is now an informational.
- A command procedure CXX$PRODUCT_REMOVE has been added to the HP C++
V7.1 kit for OpenVMS Alpha. This procedure allows you to remove the HP
C++ compiler product. It performs the equivalent of a PCSI PRODUCT
REMOVE command.
You are required to disable the product license
before issuing the command procedure to prevent a compilation from
interfering with the delete process. If the compiler has been
installed as a shared image, the command procedure will uninstall the
image. The command procedure takes no parameters and can be run as
follows:
$ @SYS$SYSTEM:CXX$PRODUCT_REMOVE
Do you wish to proceed with removing HP C++ <No>? Yes [Ret]
|
Enter "Yes" to remove the compiler from your system.
- Previously, the propagation of a C++ exception out of a thread's
start routine did not result in cxxl$terminate() being called. A
partial solution for the problem (a back-port of the solution on
OpenVMS I64) is available on OpenVMS Alpha Version V8.3 and higher. It
requires pthreads library patch VMS83A_PTHREAD-V0100.
For the new
behavior, you must link against the shareable image of the C++ Standard
Library, available with Version 7.1 and higher of the C++ Compiler for
OpenVMS Alpha (see Section 7.2), and must define the logical
CXXL$LANGRTL to point to the appropriate shareable version of the C++
Standard Library. For example:
$ define cxxl$langrtl libcxxstd
$ cxx test
$ cxxl test
$ run test
Work thread starting.
Custom terminate function called.
Work thread starting.
Custom terminate function called.
%TRACE-F-TRACEBACK, symbolic stack dump follows
image module routine line rel PC abs PC
...
$ type test.cxx
// Begin test.cxx
#include <stdlib.h> // EXIT_SUCCESS
#include <pthread.h>
#include <iostream> // cout
#include <exception> // set_terminate
#include <new> // std::bad_alloc
void terminator() {
cout << "Custom terminate function called." << endl;
abort();
}
void* work(void* param) {
cout << "Work thread starting." << endl;
set_terminate(terminator);
throw std::bad_alloc();
return 0;
}
int main (int argc, const char ** argv) {
set_terminate(terminator);
pthread_t t;
pthread_create(&t, NULL, &work, NULL);
pthread_detach(t);
sleep(2);
throw std::bad_alloc();
return EXIT_SUCCESS;
}
// End test.cxx
|
- __MEMxxx builtin functions treated length as signed.
The
builtin functions __MEMMOVE, __MEMCPY, and __MEMSET were erroneously
generating code to sign-extend the length parameter before passing it
to an OTS$ routine that interprets the length as a signed 64-bit value.
The length parameter for the standard C library routines is of type
size_t, which is unsigned int on OpenVMS. But the sign extension done
by these builtins made a length greater than or equal to 2147483648 be
seen as negative by the underlying OTS$ routines, and those routines
treat a negative length as a no-op. Although lengths so large are
unusual, they are possible and need to be supported.
It is possible that some source code exists that actually computes
negative length values, and relies on these values being treated as
no-ops. That behavior is not supported by the standard, although it is
a convention used in similar VMS library routines. With this bug fixed,
such code will most likely ACCVIO at run-time. Such code needs to be
changed to test the length value explicitly to determine if it is in a
range that should be ignored, and either bypass the call or use a
length of zero - that is the only way to assure correct operation.
Note that the
<string.h>
header may (under __NONAMESPACE_STD) define macros to replace
invocations of the standard C functions memmove, memcpy, and memset by
invocations of these builtins. And these functions are also recognized
as intrinsics, so even without the macros the compiler will treat them
as builtin functions. Therefore, just recompiling a module will
introduce the effect of this fix, even if linking against the same
version of the CRTL. The only reliable way to ensure that a negative
length value will be treated as a no-op instead of a large positive
value when using these standard C functions is to modify the source to
test the value explicitly - that's what the standard-conforming
behavior requires.
Finally, note that this bug is not present in either the builtin or
CRTL implementions of these functions on OpenVMS I64, so this fix makes
the Alpha behavior match the I64 behavior.
- Previously, the compiler was incorrectly deducing the template
argument type as a const-qualifed (or volatile-qualified) type, instead
of as an unqualified type, when deducing from a const- (or
volatile-)qualified array type.
template <class T>
void foo(const T &value) { }
void f(void) {
const int i[3] = { 1, 2, 3 };
foo(i);
}
|
The compiler previously deduced T to be of type "const int[3]"
while it now deduces it to be of type "int [3]". This affects
-model ansi
compilations only, since
-model arm
compilations do not mangle in the template argument type.
- A problem has been corrected in the implicit include processing.
The implicit inclusion will no longer select files such as "
.C
" or "
.CXX
" (where these files have no file name portion).
6.2 Restrictions in Version 7.1
- Debug is not able to break on a non-local constructor in a
shareable image because of the order in which images start up on
OpenVMS. Specifically, the debugger is loaded and begins to run only
after the initialization of all the shareable images but before the
initialization of the main image. For example, in the following program
where the constructor
foo::foo()
has been built into a shareable image, the constructor runs before the
debugger has been activated. As a result, there is no way to break on
this constructor, even if you attempt to signal the debugger.
$ cxx/debug/noopt static_ctor_main.cxx
$ cxx/debug/noopt static_ctor.cxx
$ cxx_link/share/debug static_ctor/opt
$ cxx_link/debug static_ctor_main/opt
$ define STATIC_CTOR DISK$:[DIR]STATIC_CTOR.exe
$ run static_ctor_main
foo::foo() <----- ctor has already been executed
OpenVMS Debug Version VX.X-xxx
%DEBUG-I-INITIAL, Language: C++, Module: STATIC_CTOR_MAIN
%DEBUG-I-NOTATMAIN, Type GO to reach MAIN program
--------- static_ctor_main.cxx -----------
main()
{
}
---------- static_ctor.cxx -------------
extern "C" int printf(const char *,...);
#define SS$_DEBUG 1132
extern "C" int lib$signal(unsigned int cond);
class foo {
public:
foo() {
printf("foo::foo()\n");
#if SIG
lib$signal(SS$_DEBUG);
#endif
}
~foo() { }
};
foo x;
-------- static_ctor_main.opt----------
NAME = main
IDENTIFICATION = "V1.001"
GSMATCH = LEQUAL, 1, 001
static_ctor_main.obj
static_ctor/share
-------- static_ctor.opt ------------
NAME = static_ctor
IDENTIFICATION = "V1.001"
GSMATCH = LEQUAL, 1, 001
static_ctor.obj
|
- The C++ compiler incorrectly mangles top-level cv-qualifiers into
function signatures in the default object model (/MODEL=ARM on OpenVMS,
-model arm
on Tru64 UNIX). For example:
file.cxx
-----
void f(const int);
void f(int) {}
> cxx/noobj file.cxx
void f(int) {}
.....^
%CXX-W-NOTQUACOMPREDEC, declaration is not qualifier compatible with
"void f(const int)" (declared at line 1)
at line number 2 in file DEVICE$:[DISK]FILE.CXX;15
> cxx/noobj/model=ansi file.cxx
>
|
- The function _RWdestroy(ForwardIterator first, ForwardIterator
last) has been modified to destroy the objects pointed to by iterators
in the range [first, last]. Before the fix, the function was just
looping through iterators without destroying the objects.
The
functions std::uninitialized_copy(), std::uninitialized_fill() and
std::uninitialized_fill_n() have been modified to increment iterator(s)
after the call to the _RWconstruct() function. This is to make sure
that if an exception is thrown during constructing the object, only
objects that were fully constructed by the std::uninitialized_*()
function are destroyed.
7 Release Notes for the Version 7.1 C++ Standard Library
The following sections describe enhancements, changes, and restrictions
for the C++ Standard Library.
7.1 Enhancements, Changes, and Problems Corrected in Version 7.1
- While applications using the C++ library iostreams can be compiled
with the _LARGEFILE macro defined, the C++ library iostreams do not
support seeking to 64-bit file offsets. For more information on
_LARGEFILE macro see the HP C Run-Time Library Reference Manual for
OpenVMS Systems.
- This version of the C++ compiler implements C++ headers for C
Library Facilities. The <cname> headers avoid pollution of the
global namespace by defining all C names only in namespace std. (See
Stroustrup, §9.2.2 and §16.1.2.)
The <cname>
headers are located in the same text library as the C++ Standard
Library headers and template definition files:
SYS$LIBRARY:CXXL$ANSI_DEF.TLB. If you include a <cname>
header and use the /PURE_CNAME option, all C functions and types found
in that header file are declared only in namespace std, as specifed by
the C++ International Standard. Specifying the /NOPURE_CNAME option
causes <cname> headers to be handled as if the corresponding
<name.h> version had been included. That is, names are available
both in namespace std and global scope. Otherwise, the default is
/PURE_CNAME when compiling with /STANDARD=STRICT_ANSI and
/NOPURE_CNAME. Including <name.h> after including the
corresponding <cname> header brings all names declared in that
<cname> header into global namespace with
using std::name
declarations. New overloaded function signatures have been added to
several <cname> headers (see Standard §21.4, 25.4, and
26.5). These overloaded signatures have been made available when
including the <cname> header in PURE_CNAME mode. If the
__CNAME_OVERLOADS
macro is defined, the new signatures are available in both PURE_CNAME
and NOPURE_CNAME modes. Defining the
__CNAME_OVERLOADS
macro in NOPURE_CNAME mode in combination with other macros and options
(for example, /STANDARD=MS, /DEFINE=_XOPEN_SOURCE) can cause compile
errors. cmath (Standard §26.5) now provides float and long
double overloaded signatures for math functions in PURE_CNAME mode, or
in NOPURE_CNAME mode with the
__CNAME_OVERLOADS
macro defined. These added signatures could cause type ambiguity
problems or different runtime behavior in existing code. Consider these
examples:
- sin(1) is now ambiguous because overloads are provided for float,
double, and long double. A user sees the following differences, because
the argument of sin(1) is assumed to be of type int:
cout << "sin(1) = " << sin(1) << endl; // generates an error:
%CXX-E-AMBOVLFUN, more than one instance of overloaded function "std::sin"
matches the argument list:
function "std::sin(double) C"
function "std::sin(float)"
function "std::sin(long double)"
argument types are: (int)
|
- The type of the argument to an overloaded math function determines
the type of its return value and associated precision. Calls to math
functions using float or long double arguments may return less precise
or more precise values than previously. Compare the following:
- Previous compiler release:
long double ldout = sin(1.0);
ldout = 0.84147098480789650000
|
- Current compiler release:
long double ldout = sin(1); // type int argument - ambiguous
long double ldout = sin(1.0l); // type long double argument
ldout = 0.84147098480789650000
long double ldout = sin(1.0f); // type float argument
ldout = 0.84147095680236816000
|
Signatures have been added in the cstring, cwchar, and cstdlib
header files for the following functions:
- cstring: (Standard §21.4) strchr, strpbrk, strrchr, strstr,
memchr
- cwchar: (Standard §21.4) wcschr, wcspbrk, wcsrchr, wcsstr,
wmemchr
- cstdlib: (Standard §25.4) bsearch, qsort, (§26.5) abs,
div
The added signatures could cause problems in existing code. For
example, because char* strchr(const char*, int) now has overloads const
char* strchr(const char*, int) and char* strchr(char*, int), the
following code does not work:
#include <cstring>
void f(char*) {;}
int main() {
f(strchr("abc",1)); // strchr returns a const char*
return 0;
}
|
- A problem with vector container created by a constructor accepting
two input iterators has been corrected. After the fix, the constructor
populates the container with all the contents of the stream associated
with the iterator, as it should. Before the fix, the constructor would
put only the first stream record into the container. A program like the
program example in Section 3.8.3 "Iterators and I/O" in Stroustrup's
C++ Programming Language, 3rd edition, now generates the
correct result.
- A bug in codecvt<wchar_t,char,mbstate_t>::encoding()
specialization of the
codecvt::encoding()
member function has been fixed. The function used to return 0
regardless of the encoding established by the facet while, according to
section 22.2.1.5.2, p7 of the C++ standard, it should return -1 if the
encoding is state-dependent, a constant number of characters needed to
produce a wide character (as for a single-byte character set), and 0
otherwise (as for a multibyte character set).
For example, after
the fix, the program below generates the following output:
Before the fix, it would generate the following output:
x.cxx
-----
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <locale>
#include <iostream>
using namespace std;
int main()
{
codecvt_byname<wchar_t,char,mbstate_t>* p;
// C locale
p = new codecvt_byname<wchar_t,char,mbstate_t>("");
cout << p->encoding() << endl;
// single-byte locale
p = new codecvt_byname<wchar_t,char,mbstate_t>("en_US.ISO8859-1");
cout << p->encoding() << endl;
// multibyte locale
p = new codecvt_byname<wchar_t,char,mbstate_t>("ja_JP.SJIS");
cout << p->encoding() << endl;
}
|
- codecvt<wchar_t,char,mbstate_t>::max_length() specialization
of the
codecvt::max_length()
member function has been modified to return MB_CUR_MAX for the encoding
established by the facet instead of MB_LEN_MAX. While not a bug,
MB_CUR_MAX is a more accurate return value for the
max_length()
function, and this is what the function returns in other
implementations of the C++ Standard Library, including recent versions
of Rogue Wave library.
- To synchronize access to the reference count in the reference
counting implementation of std::string class, the C++ Standard Library
uses atomic instructions. Version 7.1 of the compiler introduces an
alternate synchronization mechanism based on the pthread mutex embedded
in _RWstring_ref_rep class and using the TIS interface.
A
mutex-based synchronization can provide better performance in
configurations with slow memory access, especially, when an application
is not threaded and TIS mutex blocking becomes a no-op. Also, a
mutex-based synchronization is more robust in some situations. For
example, when an application fails to provide proper high-level
synchronization when operating on objects of std::string class in
different threads, a mutex-based synchronization might allow the
application to "survive". However, HP does not recommend relying on
this feature. To enable mutex-based synchronization in std::string
class, a program should be compiled with the
__USE_EMBEDDED_PTHREAD_MUTEX macro defined. Additionally, a program
should be using a nopreinstantiation version of the C++ Standard
Library, which assumes compiling with the __FORCE_INSTANTIATIONS macro
defined and linking /NOPREINST. Objects generated by compiling with
the __USE_EMBEDDED_PTHREAD_MUTEX macro defined are not
binary-compatible with objects generated by compiling without the macro
defined, and should not be linked in the same application. This is also
true for dynamically loaded C++ libraries. That is, with respect to the
__USE_EMBEDDED_PTHREAD_MUTEX macro, all dynamically loaded libraries
and the main executable should be compiled the same way. A vendor whose
library is built with the macro defined should notify their users to
also compile with the macro defined (and use nopreinstantiation C++
Standard Library).
- A bug in codecvt<char,char,mbstate_t>::encoding()
specialization of the codecvt::encoding() member function has been
fixed. The function used to return -1 while, according to section
22.2.1.5.2, p7 of the C++ standard, it should return 1. The fix makes
sure that the specialized function returns 1.
- A bug in codecvt<char,char,mbstate_t>::in(), out(), and
unshift() specialization of the codecvt member functions has been
fixed. These functions used to return codecvt_base::error while,
according to section 22.2.1.5.2 of the C++ standard, they should return
codecvt_base::noconv. The fix makes sure that the specialized functions
return codecvt_base::noconv.
- The codecvt<char,char,mbstate_t>::always_noconv() function
has been modified to return
true
to comply with section 22.2.1.5.2, p8 of the C++ standard. In previous
complier releases, this function used to return
false
.
- To be consistent with Library Issue 103, the reverse_iterator
typedef in set and multiset has been changed to
_RWrep_type::const_reverse_iterator.
- Because of a bug in the C++ standard library, it was impossible to
define and use a user-defined facet. For example, the following program
would not compile. This has been fixed.
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <locale>
struct foo : std::locale::facet
{
static std::locale::id id;
};
std::locale::id foo::id;
int main()
{
std::use_facet<foo>(std::locale());
return 0;
}
|
- A problem with formatting a hexadecimal number using the
ios_base::internal adjustfield manipulator has been corrected. For
example, after the fix, the program below outputs the correct string:
"0x0021". Before the fix, it would output: "000x21".
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <iostream>
#include <iomanip>
int main()
{
std::cout << std::hex << std::showbase << std::setfill('0')
<< std::setw(6) << std::internal << 33 << '\n';
}
|
- The library previously used the same storage for
iarray
(an array of long integers) and
parray
(an array of pointers to void), manipulated by the
iword()
and
pword()
member functions, respectively, of class std::ios_base. This has been
corrected so that
iarray
and
parray
now use separate storage. For example, after the fix, the following
program outputs the correct result:
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <iostream>
int main()
{
int const index = std::ios_base::xalloc();
std::cout.iword(index) = 42L;
std::cout << "iword=" << std::cout.iword(index) << std::endl;
std::cout.pword(index) = 0;
std::cout << "iword=" << std::cout.iword(index) << std::endl;
}
|
Correct result:
Before the fix, it would output:
- To comply with 21.2 - String classes [lib.string.classes] - in the
C++ standard, declarations of the
std::getline()
function operating on basic_istream have been moved from
<istream> to <string>. Accordingly, the definition of the
std::getline()
function operating on basic_istream and accepting the delim
parameter has been moved from <istream.cc> to <string.cc>.
This change is visible only when using the standard iostreams.
- A problem has been corrected with the assignment operator of the
tree container not storing the comparison object of the container being
copied into the target container.
The tree container is the
underlying container for the map and set STL containers. Because of
this problem, after assigning one STL container object to another, the
target container would continue to use the comparison object it was
using before the assignment. It violates section 23.1.2 - Associative
containers [lib.associative.reqmts] of the C++ standard which states:
- When an associative container is constructed by passing a
comparison object the container shall not store a pointer or reference
to the passed object, even if that object is passed by reference. When
an associative container is copied, either through a copy constructor
or an assignment operator, the target container shall then use the
comparison object from the container being copied, as if that
comparison object had been passed to the target container in its
constructor.
- The HP C++ library defines std::ostream_iterator as the following:
template <class T, class charT = char, class traits = char_traits<charT> >
class _RWSTDExportTemplate ostream_iterator :
public iterator<output_iterator_tag,T,_TYPENAME traits::off_type,T*,T&>
|
However, section 24.5.2 - Template class ostream_iterator
[lib.ostream.iterator] of the C++ standard defines
std::ostream_iterator as the following:
template <class T, class charT = char, class traits = char_traits<charT> >
class ostream_iterator:
public iterator<output_iterator_tag, void, void, void, void>
|
HP C++ Version 7.1 introduces the macro __COMPLY_WITH_24_5_2. When
compiled with this macro defined, the library provides the
standard-compliant definition of std::ostream_iterator. Note that
defining the __COMPLY_WITH_24_5_2 macro changes the types defined by
std::ostream_iterator, namely: value_type, difference_type, pointer,
and reference. Because of changing types, it is a good idea to make
sure that if the macro is defined, it is defined consistently in your
application and in the libraries the application is using. For
example, consider the following program:
x.cxx
-----
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <iterator>
#include <iostream>
#include <typeinfo>
using namespace std;
int main() {
cout << typeid(ostream_iterator<char>::value_type).name() << endl;
cout << typeid(ostream_iterator<char>::difference_type).name() << endl;
cout << typeid(ostream_iterator<char>::pointer).name() << endl;
cout << typeid(ostream_iterator<char>::reference).name() << endl;
}
|
When compiled without the __COMPLY_WITH_24_5_2 macro defined,
x.cxx
gives:
When compiled with the __COMPLY_WITH_24_5_2 macro defined, it gives:
- The default constructor of class template
std::istream_iterator<class T, class charT, class traits, class
Distance> used to initialize stream data members with the address of
std::cin regardless of the charT template argument. As a result, this
class template could not be instantiated on charT template arguments
other than 'char'. For example, compilation of the following program
would result in a BADINITTYP error. This has been corrected.
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <iterator>
#include <istream>
std::istream_iterator<int, wchar_t> x;
|
- The
min()
and
max()
functions in the <long long> specialization of the
std::numeric_limits class template used to return the wrong values.
Specifically, they returned the same (32-bit) values that the
min()
and
max()
functions in the <long> specialization are returning. This has
been corrected.
Consider the following sample program
x.cxx
:
x.cxx
-----
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <limits>
#include <iostream>
using namespace std;
main() {
cout << "max() = " << numeric_limits<long long>::max() << endl;
cout << "min() = " << numeric_limits<long long>::min() << endl;
cout << "max() = " << numeric_limits<unsigned long long>::max() << endl;
}
|
Without the fix,
x.cxx
gives:
max() = 2147483647
min() = -2147483648
max() = 4294967295
|
After the fix, it gives:
max() = 9223372036854775807
min() = -9223372036854775808
max() = 18446744073709551615
|
- The C++ Standard Library header <vector> was modified to
expose std::vector<bool> overloads of relational operators only
when compiling with the __DECFIXCXXL1941 macro defined. These overloads
make it impossible to use relational operators on
vector<bool>::iterator types; see the code example below. That
the current C++ standard lists these overloads (section 23.2.5 - Class
vector<bool> [lib.vector.bool]) is considered to be a defect in
the standard. Some other implementations of STL do not provide these
overloads.
With std::vector<bool> overloads of all the
relational operators removed, the following program compiles. Before
the change, it would not compile.
#include <iterator>
#include <vector>
class D : public std::reverse_iterator<std::vector<bool>::iterator> {
};
int main(void)
{
D x, y;
if ( std::operator== <std::vector<bool>::iterator>(x,y) )
return 0;
if ( std::operator!= <std::vector<bool>::iterator>(x,y) )
return 0;
if ( std::operator< <std::vector<bool>::iterator>(x,y) )
return 0;
if ( std::operator<= <std::vector<bool>::iterator>(x,y) )
return 0;
if ( std::operator> <std::vector<bool>::iterator>(x,y) )
return 0;
if ( std::operator>= <std::vector<bool>::iterator>(x,y) )
return 0;
return 1;
}
|
7.2 The C++ Standard Library in the Form of a Shareable Image
Starting with C++ Version 7.1, the compiler kit provides linker options
files and a CXXL$BUILD_SHARED_LIBCXXSTD_IMAGES.COM DCL procedure for
building the following shareable images:
- LIBCXXSTD.EXE - model ARM preinstantiation library
- LIBCXXSTD_NOINST.EXE - model ARM noinstantiation library
- LIBCXXSTD_MA.EXE - model ANSI preinstantiation library
- LIBCXXSTD_MA_NOINST.EXE - model ANSI noinstantiation library
Notice that the filenames of the shareable images are the same as the
filenames of their OLB counterparts in SYS$LIBRARY.
7.2.1 How To Create Shareable Images
The compiler installation procedure places
CXXL$BUILD_SHARED_LIBCXXSTD_IMAGES.COM into the SYS$SYSTEM directory
and places the linker options files into SYS$LIBRARY. The filenames of
the options files are the same as the filenames of the images they
create, but with a CXXL$ prefix. For example, the options file for the
model ARM preinstantiation library image LIBCXXSTD.EXE is
CXXL$LIBCXXSTD.OPT. The compiler installation procedure does not invoke
CXXL$BUILD_SHARED_LIBCXXSTD_IMAGES.COM.
To build shareable images, invoke
CXXL$BUILD_SHARED_LIBCXXSTD_IMAGES.COM as follows:
@SYS$SYSTEM:CXXL$BUILD_SHARED_LIBCXXSTD_IMAGES.COM
The procedure creates four shareable images from their respective OLB
counterparts. It accesses the OLB libraries using the SYS$LIBRARY
logical name and places the images into the same directory on the
SYS$LIBRARY search list where the OLB libraries are located. The
procedure does not insert images into IMAGELIB.OLB.
7.2.2 Linking Against the C++ Standard Library Shareable Image
There are two ways of using the C++ Standard Library in the form of a
shareable image: with and without inserting the library image into
SYS$LIBRARY:IMAGELIB.OLB.
Note that the following procedures are the same for object models ANSI
and ARM, and for linking against the preinstantiation and
noinstantiation libraries. Only name of the C++ Standard Library image
is different.
When linking against the OLB library, the CXXLINK utility automatically
chooses the library based on the /MODEL and /NOPREINST qualifiers, so
that the selection of OLB library is transparent. When linking against
the shared C++ Standard Library, you must specify the correct Standard
Library image yourself.
The following examples use the model ARM preinstantiation library
LIBCXXSTD.EXE. This is the default library; it is used when a program
is compiled without any /MODEL qualifier and without the
__FORCE_INSTANTATIONS macro defined.
7.2.2.1 Linking Without the C++ Standard Library Image in IMAGELIB.OLB
The following example shows how to link without the C++ Standard
Library image in IMAGELIB.OLB:
cxx foo.cxx
cxxl foo.obj, sys$input:/opt
LIBCXXSTD.EXE/share
^Z
define LIBCXXSTD disk:[directory]LIBCXXSTD.EXE;
run foo
|
7.2.2.2 Linking With the C++ Standard Library Image in IMAGELIB.OLB
This is probably the most convenient way of linking against shared C++
Standard Library. However, there are several restrictions:
- Model ARM and model ANSI images cannot both be inserted into
IMAGELIB.OLB because there are some symbols with the same names in the
two libraries.
- Preinstantiation and noinstantiation images cannot both be inserted
into IMAGELIB.OLB. This is because the noinstantiation library is a
subset of the preinstantiation library. Therefore, inserting a
preinstantiation library image into IMAGELIB.OLB will break linking
/NOPREINST because library instantiation symbols will be resolved from
IMAGELIB.OLB instead of from the user instantiations in the repository.
So inserting a C++ Standard Library image into IMAGELIB.OLB is suitable
for a site where only a single object model is used and where linking
is consistently done against either preinstantiation or noinstantiation
library.
In order to link against the C++ library shareable image in
IMAGELIB.OLB, create an empty LIBCXXSTD*.OLB library and make CXXLINK
use it instead of the OLB library in SYS$LIBRARY. You can do this by
defining a logical name CXX$LINK_LIBCXXSTD_DIR pointing to the location
where the empty OLB library resides. For example, for the model ARM
preinstantiation library:
libr/create disk:[directory]LIBCXXSTD.OLB
define CXX$LINK_LIBCXXSTD_DIR disk:[directory]
|
Note that while the name of the logical remains the same, the name of
the OLB library it points to depends on the object model and the
preinstantiation/noinstantiation library. For example, for model ANSI
it would be LIBCXXSTD_MA.OLB; for the model ARM noinstantiation
library, it would be LIBCXXSTD_NOINST.OLB, and so on.
With the C++ library shareable image in IMAGELIB.OLB, you can link
against it without having to explicitly specify it on the CXXLINK
command. For example:
cxx foo.cxx
cxxl foo.obj
run foo
|
7.2.3 Restrictions
7.2.3.1 Overriding Global Operators new and delete
As the C++ User's Guide indicates, programs overriding global new and
delete must be linked /NOSYSSHARE. Such programs will have to continue
linking /NOSYSSHARE against the OLB libraries.
7.2.3.2 Mixing OLB and Shared C++ Library in the Same Process
Mixing dynamically loaded libraries linked against OLB and the shared
C++ Standard Library in the same process is not supported. Also, a main
executable must be linked against the same flavor (either .OLB or .EXE)
of the C++ Standard Library against which the libraries it dynamically
loads are linked. Violating this restriction can result in
unpredictable behavior.
7.3 Restrictions in Version 7.1
This section describes problems you might encounter when using the
current release of the C++ Standard Library with the HP C++ compiler.
Where appropriate, workarounds are suggested.
7.3.1 Using the C++ Libraries in Microsoft Standard Mode
When compiled /STANDARD=MS, the following restrictions apply:
-
typeid(__int64).name()
returns
__int64
instead of
long long
.
- Header
<new>
does not declare operator
delete[](void*, void*)
.
- Header
<new>
does not declare type
new_handler
.
-
mem_fun
function objects from
<functional>
cannot be used with a void first template argument. For example, the
following program does not compile in Microsoft mode:
template <class S, class T>
struct mem_fun_t
{
S (T::*pmf)();
S operator()(T* p) { return (p->*pmf)(); }
};
struct X {};
void (mem_fun_t<void, X>::*pf) (X*) = &(mem_fun_t<void, X>::operator());
|
- It is impossible to create a list container of elements of type
size_t
using the list(size_type) constructor. For example, the following
program does not compile in Microsoft mode:
#include <stddef.h>
#include <list>
std::list<size_t> x(1);
|
- Using
ostream<<
and
istream>>
operators with variables of type
[unsigned] __int64
results in undefined linker symbols.
- Using
vector<bool>
iterators results in compilation errors. For example, the following
program does not compile in Microsoft mode:
#include <vector>
void f()
{
std::vector<bool> x;
x.insert(x.begin(), true);
}
|
7.3.2 Other Restrictions
Other Standard Library restrictions in this release are:
- The std::use_facet<FACET> function template can be
instantiated only on facets having a default constructor.
8 Release Notes for the V6.5 C++ Compiler
The following sections describe enhancements, changes, and restrictions
for the C++ compiler environment.
8.1 Enhancements, Changes, and Problems Corrected in 6.6-???
- The compiler was not correctly handling certain unusual loops
within a switch statement. (11108)
- For a template specialization, the compiler mangles names
differently depending on whether specialization is used. For example,
if a function having parameter of
vector<bool>
type is the only entity in compilation unit referencing
vector<bool>
, function name will be mangled differently depending on whether the
function actually uses its
vector<bool>
argument in the function body.
In order to eliminate this
dependency, a dummy function was added to the C++ standard library
header
<vector>
, as the following:
#ifndef _RWSTD_NO_CLASS_PARTIAL_SPEC
#if defined(__DECCXX) && !defined(__DECFIXCXXL1760)
inline void __function_to_use_vector_bool()
{
vector<bool> x;
}
#endif
#endif
|
This function ensures, that
vector<bool>
specialization of
vector<>
is always treated as "used" in any program that
#include
s the
<vector>
header. (L1760)
- Changes to
<time.h>
and
<time.cc>
This kit supplies updated versions of DECC$RTLDEF.TLB (C library
files) and CXXL$ANSI_DEF.TLB (C++ standard library header files) that
address various issues in an upward-compatible fashion. However, there
are changes to the conditionalization of the layout and member names
for the type "struct tm" (from
<time.h>
), as well as conditionalization of the implementation of the gmtime()
and localtime() functions [and their reentrant variants gemtime_r() and
localtime_r()] to be used, that might affect some programs. As a
general precaution to minimize possible problems, it is recommended
that if any source module that #includes either
<time.h>
or
<time>
is recompiled, all modules in the application that
#include
either of those headers should also be recompiled. Note that the C++
header
<time>
#includes
<time.cc>
(when /NOIMPLICIT_INCLUDE is in effect). The C++ header
<time>
in itself is not affected. The purpose of the changes is to
address an interrelated set of functionality, user-namespace, and
binary compatibility issues in a way that will be consistent between C
and C++ going forward with new versions of the compilers and run-time
libraries.
- Functionality:
The C++ standard library's "time_put" time
formatting facet supports timezone formatting from a value of type
"struct tm". Therefore, C++ compilations need to ensure that the
implementations of localtime() and gmtime() selected [and their
reentrant _r variants] are the ones that support timezone information.
One specific functionality bug addressed by the changes is that some
versions of the header provided two additional members for timezone
information in the declaration of struct tm for use by C++, but did not
cause the timezone-aware implementations of the time functions to be
selected under those same conditions. This could lead to use of
uninitialized data, or an access violation when the user code then
tried to access the additional members of struct tm.
- User-namespace:
The members of struct tm that support timezone
information are not part of the C standard. They are actually a common
extension from BSD Unix, with conventional names of tm_gmtoff and
tm_zone. Although the C standard allows an implementation to define
additional members in struct tm, it does not reserve the prefix "tm_"
for that purpose. So a strictly-conforming C program could #define
tm_zone as a macro expanding into something that would cause a syntax
error when encountering tm_zone as the name of a member in a struct
type. Or conversely, a program compiled with the strict C namespace in
effect could refer to the members by their BSD names, without getting a
diagnostic, and infer that the names were defined under the C standard.
Previous versions of
<time.h>
handled this issue by suppressing the declaration of these two members
entirely under the strict C namespace (_ANSI_C_SOURCE). This new
version (when used with C++ versions *newer* than V6.5-042) defines
struct tm such that the timezone members have names in the namespace
reserved to the implementation when the strict C namespace is in
effect: instead of tm_gmtoff and tm_zone it uses __tm_gmtoff and
__tm_zone (which is the convention on Tru64 and Linux systems). Because
the
<time.cc>
header contains source code that refers to the "tm_zone" member of
struct tm, that code had to be conditionalized to use "__tm_zone" when
the strict C namespace is in effect (and continue to use the BSD names
when it is not in effect).
- Binary compatibility:
Previous versions of
<time.h>
that suppressed the two non-standard members in struct tm under the
strict C namespace resulted in potential binary incompatibility between
modules compiled with/without the strict C namespace. The practice in
the new
<time.h>
of just changing the names of the members to meet the requirements of
the strict namespace makes modules compiled with either setting binary
compatible, but it introduces potential for binary incompatibility with
pre-existing object modules that cannot be recompiled from source.
Note that modules compiled with different sizes for struct tm only have
potential for encountering a problem; in typical usage no problem will
occur because struct tm is not often embedded within a struct or used
as an element of an array, with that struct or array then shared beteen
compilation units. Typical code using struct tm only accesses it
locally through the pointer returned by the library functions, or
contructs a local instance that is then passed by address to library
functions. Those kinds of uses are not affected by a size difference
that occurs between different compilations. But to help deal with
possible situations where a real binary compatibility problem is
encountered, and it is not feasible to recompile all of the modules
involved, the struct declaration in
<time.h>
can be forced to the short version (without the timezone members) by
defining the macro "_TM_SHORT" before the header is included.
Similarly, the declaration can be forced to the long version by
defining "_TM_LONG". These macros will override any other
conditionalization based on the version of the CRTL and the compiler,
and produce a struct tm declaration that is of the specified size.
As described above, the use of implementation-reserved identifiers
to make the size of struct tm uniform regardless of whether or not the
strict C namespace is in effect is conditionalized to the new C++
compiler. These headers are shared between C and C++ compilers, and
that change will not affect existing versions of the C compiler. Unlike
C++, the C compiler enables the strict C namespace under its
/stand=ansi89 or /stand=c99 qualifiers as well as when any of the
_XOPEN, _POSIX, or _ANSI_C standard-conformance macros are specified
(see section 1.5 Feature-Test Macros for Header-File Control in the C
Run-Time Library Reference Manual). It was felt that the risk of binary
compatibility problems would be slightly greater for C than for C++
because these options are more commonly used with C compilations than
they are with C++. However, the _TM_LONG (and _TM_SHORT) macros will
affect both compilers. If your application contains C modules that
#include
<time.h>
(as well as C++ modules that do), it is advisable to recompile those C
modules with /define=_TM_LONG to avoid possible binary incompatibility
between the C and C++ modules.
- A problem with catching a VMS exception when compiled /names=as_is
and/or with the __NEW_STARLET macro defined, has been fixed. In this
compiler release, however, the fix is available for object model ARM
only. The fix for object model ANSI is planned to be released in the
next major compiler release for OpenVMS Alpha.
For example, the
following program now outputs "caught signal". Before the fix, if
compiled /names=as_is or with the __NEW_STARLET macro defined, it would
output "missed VMS signal". (10655)
#include <chfdef.h>
#include <lib$routines.h>
#if __DECCXX_VER <= 60590042
#define cxxl$set_condition CXXL$SET_CONDITION
#endif
#include <cxx_exception.h>
#include <stdio.h>
main() {
try {
cxxl$set_condition(cxx_exception);
lib$signal(99628);
}
catch (chf$signal_array&) {
puts("caught signal");
}
catch (...) {
puts("missed VMS signal");
}
}
|
- Exception handler now preallocates some memory so exceptions can be
thrown even if no memory is available. (7881)
- The compiler was incorrectly computing the element size as zero
when the array was a typedef-ed type. This caused problems, for
instance, when whole array destruction was invloved. (10865)
- When compiled with full IEEE support (-ieee option on Tru64 and
/FLOAT=IEEE/IEEE=DENORM qualifiers on AlphaVMS), denorm_min() function
of the specialization of class template std::numeric_limits for long
double would return garbage. It has been corrected. For example, the
program below now returns the correct value: 6.47518e-4966. (L1880)
8.2 Enhancements, Changes, and Problems Corrected in 6.5-042
- Compiler assertion compiling very large programs with debug has
been eliminated. (10828)
- Compiler crash in pipeline optimization has been eliminated.
(10216)
8.3 Enhancements, Changes, and Problems Corrected in V6.5-041
- A program using standard iostreams for non-standard __int64 and
long long datatypes could not be compiled in strict ansi mode. This
restriction has been eliminated. (L1823)
- Eliminate buffer overrun in CXXLINK. (10633)
- Eliminate incorrect diagnostic on using declaration of template
base class. (10513)
- Eliminate compiler assertion when using a complex expression in
the condition of a for loop. (10678)
8.4 Enhancements, Changes, and Problems Corrected in V6.5-040
- Problem looking up non-dependent names that are static has been
corrected. (10458)
- Compiler crash when compiling in strict ANSI mode has been
eliminate. (10460)
8.5 Enhancements, Changes, and Problems Corrected in V6.5-039
- Eliminated incorrect unreachable diagnostic introduced in the
V6.5-034 compiler. (10307)
- Corrected template argument deduction bug. (10352)
- Corrected bad debug information generated for importing typedef
into namespace. (10280)
8.6 Enhancements, Changes, and Problems Corrected in V6.5-038
- Improve compilation speed by eliminating the opening and closing
of header files that have include once preprocessor guards. (8786)
8.7 Enhancements, Changes, and Problems Corrected in V6.5-036
- Inappropriate inaccessible member diagnostic has been eliminated.
(10215)
- Inappropriate diagnostic when initializing data members of a class
template has been eliminated. (10238)
- Compiler assertion when processing memcpy has been eliminated.
(10137)
- Bad runtime typeinfo (RTTI) has been corrected. (10156)
8.8 Enhancements, Changes, and Problems Corrected in V6.5-035
- Implemented "/first_include" which includes the specified file
before processing sources. (5434,10064)
- Eliminate compiler assertion for catch blocks with control flow
that will not reach the end of the block. (9946,10046)
- Do not issue a warning if a template parameter is not used in
signature of template if it is possible to use the template. (9909)
8.9 Enhancements, Changes, and Problems Corrected in V6.5-034
- Implemented "/main=posix_exit" which causes main to call
__posix_exit instead of exit when returning. (9724)
- Eliminate compiler assertion for catch blocks with control flow
that will not reach the end of the block. (9946)
- Eliminate compiler assertion for code with embedded newlines.
(9962)
8.10 Enhancements, Changes, and Problems Corrected in V6.5-033
- Compiler crash when compiling member function "-g" has been
eliminated. (9769)
- Incorrect handling of memmove of overlapping strings with
"/arch=ev56" or later has been corrected. (9791)
8.11 Enhancements, Changes, and Problems Corrected in V6.5-032
- Compiler crash in exception support has been eliminated. (9743)
- Incorrect this pointer when calling a virtual function from a
constructor has been corrected. (9751)
- Compiler crash in exception support has been eliminated. (9756)
8.12 Enhancements, Changes, and Problems Corrected in V6.5-031
- Compiler crash when using
goto
statements has been eliminated. (9666)
- Compiler crash when using multi-dimensional arrays has been
eliminated. (9675)
- Implemented prototype for
pragma include_directory
. (9677)
- Undefined symbols when using RTTI information for long long types
in model ANSI has been corrected. (9687)
- Incorrect error about using declaration has been eliminated. (9693)
- Runtime crash when using exceptions and optimization has been
eliminated. (9697)
8.13 Enhancements, Changes, and Problems Corrected in V6.5-030
- Incorrect informational about partially overridden virtual
function has been eliminated. (9343)
- Compiler hang when generating
XREF
cross reference information has been eliminated. (9561)
8.14 Enhancements, Changes, and Problems Corrected in V6.5-029
- std::deque<>::erase(iterator position) function has been
corrected to return end() if after erasing the element the container is
empty. (L1686)
- Compilation errors when compiling
builtins.h
in
ms
mode have been eliminated. (9588)
- Problem with
dynamic_cast
using class defined in shared images has been corrected. (9603)
- Error message for
utimes
function prototype when including
time.h
has been eliminated. (9609)
8.15 Enhancements, Changes, and Problems Corrected in V6.5-028
- Code generation problem resulting in failure of virtual function
override when using -nortti has been corrected. (9575)
- Fix buffer overrun in ostrstream and strstream. (L1684)
- Incorrect "extern inline function was referenced but not defined"
diagnostic has been eliminated. (8862)
8.16 Enhancements, Changes, and Problems Corrected in V6.5-026
- Code generation problem resulting in virtual function call to
member function from class of same name, but in a different namespace,
has been corrected. (9566)
8.17 Enhancements, Changes, and Problems Corrected in V6.5-024
- Release notes updated to clarify version compatibility.
8.18 Enhancements, Changes, and Problems Corrected in V6.5-021
- Compiler crash in optimizer has been eliminated. (9532)
- Missing initialization of static variables corrected. (9530)
- Compiler crash when using listing files with macro expansion and
very long lines has been eliminated. (9518)
- Optimization problem with memcpy intrinsic has been corrected.
(9512)
- When called with a null string, the basic_stringbuf::str(const
string_type& str) method was not setting the underlying character
buffer to zero length. It has been fixed. [10.1674]
-
<new>
and
<new.hxx>
library headers have been modified to suppress "support for placement
delete is disabled" informational message at the point of declaration
of nothrow version of operator delete and nothrow version of array
delete. [10.1675]
8.19 Enhancements, Changes, and Problems Corrected in V6.5-020
- Improved compiler optimization resulted in reduced abstraction
penalty in Stepanov benchmark.
- When called with a null string, the basic_stringbuf::str(const
string_type& str) method was not setting the underlying character
buffer to zero length. It has been fixed. [10.1674]
-
<new>
and
<new.hxx>
library headers have been modified to suppress "support for placement
delete is disabled" informational message at the point of declaration
of nothrow version of operator delete and nothrow version of array
delete. [10.1675]
8.20 Enhancements, Changes, and Problems Corrected in Version 6.5
- Improved conformance to the C++ International Standard
- Improved code optimization
- Retirement of CFRONT dialect announced
- The library header and template definition files have been modified
to compile in STRICT_ANSI mode with the Version 6.5 compiler.
Modification was necessary because Version 6.5 enforces more stringent
language rules in some cases than previous versions. [10.1649]
- The
deallocate()
member function of the allocator class in the
<memory>
header has been modified so that
operator delete
is not called with the null pointer. Although calling
delete
with the null pointer is legal in HP C++, the Third Degree tool
on Tru64 UNIX issues a warning when such a call is made. The
modification was made to avoid the warning. [10.1657]
8.21 Restrictions and Known Problems in Version 6.5
- If you compile and link with /MODEL=ANSI and then try to use the
VMS debugger to access members of a virtual base class, you might see
this error:
%DEBUG-E-INTERR, debugger error in DBG$GET_BASE_CLASS_OFFSET: can't find __bptr
or session corruption
|
The error is displayed because the debugger has not yet been
updated to understand the new internals of the ANSI object model.
- Comparing an object of the Standard or Class Library fstream class
with the null pointer causes an ambiguity compilation error.
For
example, the following program produces compilation error:
#include <fstream.h>
int foo()
{
fstream fs;
if (fs == NULL ) return 0;
return 1;
}
$ cxx/noobj x.cxx
if (fs == NULL ) return 0;
.........^
%CXX-E-AMBIGUOUSOPRFUN, more than one operator "==" matches these operands:
built-in operator "pointer == pointer"
built-in operator "pointer == pointer"
operand types are: fstream == int
|
The workaround is instead of comparison with the null pointer which is
based on
operator void *
() use
operator!
(). For example, if the program is modified as following:
it compiles cleanly. [L1710]
- Tight coupling of any of std::cout, std::cerr and std::clog by
making them use the same strstreambuf object may cause a deadlock in a
multithreaded application. For example, the following program may
deadlock:
ofstream log("test.log");
cout.rdbuf(log.rdbuf());
cerr.rdbuf(log.rdbuf());
...
writing to cerr from different threads
...
|
The workaround is to untie the streams before tight coupling them, as
the following:
ofstream log("test.log");
cerr.tie(0);
cout.rdbuf(log.rdbuf());
cerr.rdbuf(log.rdbuf());
|
Note, that the current implementation of the C++ Standard Library ties
all four standard streams (the C++ standard only requires, that cin is
tied to cout). [L1707]
9 Release Notes for the V6.3 C++ Compiler
9.1 Enhancements, Changes, and Problems Corrected in Version 6.3
- In
strict_ansi
mode, the name of a class is now entered as a member of itself, as
required by clause 9 (para 2) of the Standard; this behavior is
implemented more or less as an implicitly declared member typedef and
might cause some existing programs to fail. For example:
namespace std {
class iterator {};
}
struct tree
{
struct iterator {};
struct nested : public std::iterator
{
// HP C++ 6.2 and below thinks this is tree::iterator
// HP C++ ?? in strict_ansi mode thinks this is std::iterator
nested(const iterator&);
};
};
|
[6613]
- The error "incompatible parameter", issued when there is a
difference in sign between pointers, has been made discretionary. As a
result you can now reduce/increase the severity of this message or
enable/disable it using its error tag
incompatibleprm
or its error number. The same can also be done by enclosing the
offending code in #pragma. For Example, the error message for the
following program can now be controlled.
void f(unsigned int *i) {
}
void main() {
f((int *)0x05);
}
|
In addition, specifying
-std gnu
reduces the message severity to warning.
- The undocumented operator
char*()
, which converts a
String
to a pointer to
char
in the non-standard
String
class has been conditionalized with the macro
__DEC_STRING_COMPATIBILITY
so that it is available only when the program is compiled with the
macro defined.
This operator has been conditionalized beccause it
allows passing a
String
object to functions like
istream::get()
or
istream::getline()
that expect a pointer to
char
. The
String::operator char*()
returns a pointer to the "data component" of the
String
, which is a private data member inside the
String
class. When a
String
object is passed to a function expecting a pointer to
char
, the function cannot determine how much memory has been allocated by
the
String
, and the
String
cannot determine the status of the exposed pointer and the memory to
which it points. This design flaw in the non-standard
String
class has been fixed in a way that leaves the legacy functionality
available as an option. The standard ANSI C++
String
class, for example, lets users retrieve only a
const char*
, not a
char*
, to the pointer owned by the
String
. [10.1563]
- In response to user requests, CXXLINK by default now deletes the
option file SYS$DISK[]CXX_REPOSITORY.OPT that previously remained in
the user's current directory on completion of CXXLINK.
To support
this change, CXXLINK creates a unique prefix for the repository .OPT
filename in the following format:
For example, if the node name is MYNODE and the PID is 020204AEA, the file name would be MYNODE020204AEA.OPT. This option file is created in the [.CXX_REPOSITORY] directory, if one exists. If the [.CXX_REPOSITORY] directory does not exist, the
CXXLINK
. [7135]
- In Version 6.2, signaling of VMS exceptions or a C signal might
cause a program to terminate with
%CXXL-F-TERMINATE
status regardless of which VMS exception or C signal has been raised
and regardless of whether a VMS exception handler or C signal handler
has been established.
For example, the following
test1.cxx
program calling
lib$stop(SS$_MCNOTVALID)
terminates with
%CXXL-F-TERMINATE
:
#include <lib$routines.h>
#include <ssdef.h>
main() { lib$stop(SS$_MCNOTVALID); }
$ run test1
%CXXL-F-TERMINATE, terminate() or unexpected() called
%TRACE-F-TRACEBACK, symbolic stack dump follows
image module routine line rel PC abs PC
...
|
Similarly, the following
test2.cxx
program, establishing a C signal handler to catch a
SIGILL
signal, also terminates with
%CXXL-F-TERMINATE
without a signal handler being invoked:
#include <signal.h>
#include <stdio.h>
static void signal_handler(int x)
{
puts("signal arrived");
}
void main()
{
signal(SIGILL, signal_handler);
raise(SIGILL);
}
|
Version 6.3 corrects this behavior: the C++ run-time environment no
longer issues a call to
terminate()
if a VMS exception (possibly, generated by a
raise()
or
gsignal()
function arrives. Instead, the exceptions is resignaled to the next
exception handler, and the exception is handled properly. For
example, the
test1.cxx
program now correctly terminates with the OpenVMS condition value
specified in the call to
lib$stop()
as follows:
$ run test1
%SYSTEM-F-MCNOTVALID, device microcode is not valid
%TRACE-F-TRACEBACK, symbolic stack dump follows
image module routine line rel PC abs PC
...
|
And
test2.cxx
now catches the signal and terminates correctly as follows:
$ run test2
signal arrived
$
|
[7550, 8234]
- The compiler can no longer generate machine code listing when
requested to not produce object files. [8448]
- CXXLINK no longer incorrectly concatenates arguments passed to it
if such arguments are defined as logical names. [8467]
- CXXLINK now correctly handles filenames when using extended ODS-5
file parsing. [8626]
- In the ANSI object model (/model=ansi), string literals are treated
as pointers to const char (const char *). In the ARM object model
(/model=arm, the default), string literals continue to be treated as
pointers to char (char *). [8659]
- To generate the correct code to make calls to virtual functions
that return a structure, class, or union, the compiler needs the full
definition of the type returned, which is missing in the following code
fragment:
struct Cities;
struct Storage {
int x;
virtual Cities visit() = 0;
};
struct Reader : virtual Storage {
Cities visit();
};
struct T2 : Reader {
};
extern void ggg(Storage *);
void fff(void) {
Storage *s = new T2();
s-x = 23;
ggg(s);
}
|
Because the complete definition for the type Cities is missing, the
compiler cannot correctly generate the code that allows
ggg()
to call
s-visit()
for all possible definitions of the type Cities. Previous compiler
versions generated code that worked for some definitions of Cities but
not for others. The current version of generates a diagnostic when
it encounters a situation:
virtual Cities visit() = 0;
...................^
%CXX-E-INCTYPPREVIRTAB, The incomplete type Cities precludes correct
generation of the virtual table for the type Reader.
Supply the complete type definition for Cities or use
/MODEL=ANSI.
at line number ... in file ...
|
If you provide the definition of the type Cities in the module that
generates this diagnostic, the compiler can generate the correct code.
This code generation problem has been corrected in the ANSI object
model. It could not be corrected in the ARM object model because an
model incompatibility would have resulted. [8668]
- In
basic_istream
and
basic_ostream
, sentry constructor has been modified to not create the guard object
if the stream is
tie()
'ed to a stream using the same streambuf object. This eliminates the
mutex deadlock which can occur in a multithreaded application.
[10.1641]
With this modification, the following program no longer
hangs when compiled with
/define=DEADLOCK
and linked against the PTHREAD library.
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <iostream>
int main() {
ostream x(std::cout.rdbuf());
ostream y(std::cout.rdbuf());
#ifdef DEADLOCK
y.tie(&x);
#endif
y << "Hello, world" << std::endl;
return 0;
}
|
- To ensure thread safety, the
basic_string
reference count used to be protected by a mutex, which called thread locking
and unlocking routines. Performance of this class in multithreaded
applications has been improved by changing the implementation to use instead
atomic builtins (see Appendix C
in the
HP C++ User's Guide for OpenVMS Systems
).
[10.1138]
- Undefined symbols from the Standard Library no longer occur when
using the
/names=as_is
qualifier. [10.1158]
- On OpenVMS Alpha Version 7.0 and above, the Standard Library
headers now include
pthread.h
to implement multithread safety. Currently this has the side effect of
polluting the global namespace with non-reserved macro and
typedef
names such as EXCEPTION, THROW, CATCH, and so on from
<pthread_exception.h>
. This problem has been acknowledged as a defect and is being
investigated In the meantime, if you are using any of these names you
must change them or devise a workaround.
For example, if you have:
#include <vector>
#define CATCH 42 // trying to define your own macro, but get redefinition error
|
you would need to make the following change:
#include <pthread.h>
#undef CATCH // undef the pthread macro
#include <vector>
#define CATCH 42 // define your own macro
|
[10.1205]
- The string extraction operator no longer removes the extra space at
the end of the string, as in the following example
ifstream inFile("input.dat"); // input.dat contains "abc de"
inFile >> word; // read "abc"
inFile.get(ch1); // ch1 should be space, was 'd'
|
[10.1300]
- The
basic_string::find_first_not_of(charT, size_type)
function now works correctly if the string contained embedded nulls.
[10.1316]
- In Version 6.2,
basic_ostream::flush()
did not always flush the buffer if the type of the stream was an
fstream
. The buffer is now always flushed. [10.1348]
- Compiling a program in
-std strict_ansi
mode using the
basic_fstream
class no longer causes run-time seg faults or core dumps. [10.1357]
- When using the Standard Library iostreams for interactive input to
cin
from a terminal, a user may have to type more than one Ctrl D to
indicate end-of-file. [10.1413]
- The
iostreams
and
locales
are now multi-thread safe. [10.1429]
- Currently you might encounter compilation errors if you try to use
a user-defined allocator and pointer class with the STL containers.
This problem will be fixed in a future release. [10.1430]
- The Standard Library vector class now allocates space correctly for
elements greater than 1024 bytes; runtime core dumps caused by
incorrect allocation in previous versions no longer occur. [10.1459]
- The tree data structure, which sets and maps usage, has been
refined to decrease the amount of space allocated for small element
size containers. [10.1475]
- The standard library headers no longer include
pthread.h
or
tis.h
. They can therefore be used in conjunction with the macro
_PTHREAD_USE_D4
(DECthreads POSIX 1003.4a/Draft 4 (or d4) interface). [10.1540]
- The following functions have been added to
basic_filebuf
class:
bool __sync_with_device() const;
// Returns the value of _RWsync_with_device flag.
|
bool __sync_with_device(bool state);
// Sets the value of _RWsync_with_device flag and returns
// the previous value of the flag.
|
These functions are not described in the C++ standard, and are
provided as an extension in order to control the behaviour of
basic_ostream::flush()
method and
std::endl
and
std::flush
manipulators. The functions are available only on OpenVMS systems.
If
_RWsync_with_device
flag is true, and if the type of the stream is an
fstream
, the
basic_ostream::flush()
function transfers all buffered data to the device as if the C Run-Time
Library
fsync()
function were called. If
_RWsync_with_device
flag is false, it is not guaranteed that
basic_ostream::flush()
function or
std::endl
or
std::flush
manipulators will flush the data all the way to the device. To
maintain compatibility with the behaviour of the Standard Library
shipped with the HP C++ Version 6.2 compiler, the default value of
_RWsync_with_device
flag is false. The following example shows how the new functions are
used.
#ifndef __USE_STD_IOSTREAM
#define __USE_STD_IOSTREAM
#endif
#include <fstream>
#include <stat.h>
int main(void)
{
const char * filename = "file.dat";
std::ofstream testfile(filename);
struct stat buf;
// display file synchronization state
cout << "1? file synchronization state is: " <<
testfile.rdbuf()->__sync_with_device() << std::endl;
// write a record to the file
testfile << "x" << std::endl;
// check file size. Expected zero because by default endl
// manipulator does not force data to be transferred to
// the device.
stat(filename, &buf);
cout << "file size is: " << buf.st_size << std::endl;
// set _RWsync_with_device flag to TRUE and display
// previous file synchronization state
cout << "file synchronization state is: " <<
testfile.rdbuf()->__sync_with_device(1) << std::endl;
// write another record to the file
testfile << "y" << std::endl;
// check file size. Expected four because now std::endl
// will cause data to be transferred to the device.
stat(filename, &buf);
cout << "file size is: " << buf.st_size << std::endl;
return 0;
}
|
[10.1574]
9.2 Restrictions and Known Problems in Version 6.3
If you compile and link with /MODEL=ANSI and then try to use the VMS
debugger to access members of a virtual base class, you might see this
error:
%DEBUG-E-INTERR, debugger error in DBG$GET_BASE_CLASS_OFFSET: can't find __bptr
or session corruption
|
The error is displayed because the debugger has not yet been updated to
understand the new internals of the ANSI object model.
10 Release Notes for the V6.2 C++ Compiler
10.1 Enhancements, Changes, and Problems Corrected in Version 6.2A
Enhancements, changes, and problems corrected are as follows:
- The compiler now generates code that reports a static data member
as an undefined symbol at link time if the data member is referenced
but not defined. [CXXC_BUGS 6826]
- The new cxxlink design implemented in Version 6.2 to use .olbs
instead of objects is not a general solution. The default behavior of
cxxlink was changed back to using objects (
/prelink=use_object_files
) and the switch
/prelink=use_olb
was added to obtain the new mechanism. The new mechanism was simplified
for better performance. No attempt is made to try alternative link
methods if the link fails; the default link must be used. [6804, 6795,
6583, 5997]
- In the C++ Version 6.1 compiler, some objects might have their
exception unwinding information set to a negative index in the cleanup
table. This condition generates the following message:
Unexpected dtor_block_tag in do_dtor_cleanup
|
[7091]
- Fix for
raw_storage_iterator
assignment operator
In versions 6.2 and earlier, a problem in the
assignment operator for the class
raw_storage_iterator
could cause a run-time seg fault if, for example, you called the
algorithm
stable_sort()
more than once with the same container. The problem has been corrected.
[10.1284]
- Fix for
basic_string::compare()
member functions
In version 6.2, two of the
basic_string::compare()
member funrtions were throwing an exception if the length of the second
string was longer than the length of the current string. This has been
fixed so that they throw an exception only if the position the user
specifies within the second string is greater than the length of the
second string. [10.1298]
- Fix for
basic_string::resize()
A problem in the
basic_string::resize()
member function in Version 6.2 has been corrected. The incorrect
behaveior was that if two strings pointed at the same underlying
char*
, and the
resize()
function was called on one of them, and if you then you changed the
underlying string for one string, the value for the other strings value
would also be changed. [10.1287]
- Fix to string assignment operator when assigning string with
embedded nulls
A problem in the
basic_string
assignment operator prevented strings containing embedded nulls from
being copied correctly. The problem has been corrected. [10.1238]
10.2 Enhancements and Changes in Version 6.2
This release solves several problems in earlier versions of the
compiler and includes an updated OpenVMS debugger (dated 4-Mar-1999)
that solves problems with the previous debugger.
Enhancements and changes are as follows:
- Support for explicit template function arguments.
- Improved EV6 support.
- CXXLINK creates an object library.
CXXLINK now creates an
object library (CXX$LINK.OLB) in the writeable repository directory.
The object library is populated with all object files found in the
repository. CXXLINK also checks any other repository directory
listed on the command line (by use of
/REPOSITORY= qualifier
) for CXX$LINK.OLB object libraries. Each object library is then
used as input to the OpenVMS Linker through LNK$LIBRARY logical names.
In most cases, the OpenVMS Linker can resolve any missing
instantions by searching the object libraries. CXXLINK can then proceed
directly to the final link, without having to parse the linker output
looking for unresolved symbols and their matching object files.
Note the following restrictions:
- CXXLINK makes use of the OpenVMS Linker Utility's LNK$LIBRARY
logical names to reference specific object libraries as input to the
linker. If the CXXLINK command includes any form of the
/USERLIBRARY
, an informational message appears, and CXXLINK lists any required
object libraries in a linker options file.
- CXXLINK always creates and repopulates the object library. For
large repositories, this mechanism is slower than processing the linker
output and creating the linker options file listing each object file.
- CXXLINK does not create the CXX$LINK.OLB file in nonwritable
repository directories. Currently, this must be done outside of CXXLINK
by entering the command:
$ LIBRARY/CREATE non-writeable-repository-dirCXX$LINK.OLB -
non-writeable-repository-spec*.OBJ
|
- Programs that use
extern_model
can encounter problems, because object files from an object library do
not produce the same
extern_model
code that is generated when the object file is used directly.
- Modules that use
extern_model
must be listed separately as an input to the linker.
- A compiler crash caused by a label statement in a switch statement
has been corrected. [6614]
- If a user calls
VAXC$ESTABLISH
or
LIBC$ESTABLISH
in a function with a try/catch, the establish routine now works correctly. Note that C++ exception handling is disabled in these routines, as described in
C++ Exceptions, lib$establish and vaxc$establish
in
HP C++ User's Guide for OpenVMS Systems
.
- The compiler has implemented the
_poppar
builtin function and added code to convert output type for
_poppar
,
_popcnt
,
_leadz
, and
_trailz
to match contents of the new OpenVMS
builtins.h
file. See
Built-In Functions
in
HP C++ User's Guide for OpenVMS Systems
.
- SCA Version 4.6-4 and later can import C++ cross reference data
into its analysis data file. To import the data, follow these steps:
- Enter the command
CXX/XREF filename.cxx
to have compiler emit the cross reference data file
filename.xref
.
- Enter the command
SCA/IMPORT filename.xref
to generate
filename.ana
, the analysis data file.
- Because of a bug introduced in Version 6.0, the compiler emitted a
union
layout incompatible with earilier versions if the union contained
bitfield of size 8, 16, 24... up to the size of the type for the
bitfield. For example:
typedef struct size4 {
union {
unsigned ttn :8;
struct {
unsigned incr :3;
unsigned rep :5;
} v1;
} u;
} size4;
|
The size of
size4
is 4 bytes for Version 5.7, and was incorrectly set to 1 since Version
6.0. This bug is fixed in V6.2. If you have such a bitfield in your
code, you must recompile. [6567]
- A problem with
/extern_model=strict_refdef
using template static data members and inline functions has been fixed.
Previously, the compiler reported multiply defined external errors for
symbols starting with
__SDG
and
__LSG
. When creating shared images, these symbols must be put in the symbol vector of the shared image. For more information, see
Sample Code for Creating OpenVMS Shareable Images
in
HP C++ User's Guide for OpenVMS Systems
. [6506]
- A template class name can now be reused as a non-template class in
a private namespace without generating an error message. [6541]
- Mangled names beginning with
CXXL$
and
DECC$
are now included in the demangler database and can be demangled. [6458]
- More cases of unreachable code are now diagnosed. If you receive
warnings you believe to be inappropriate, please report them. [6534]
- The compiler no longer can exhaust virtual memory when computing
the addresses of deeply nested virtual base classes. [6473]
-
va_start
now handles
parmN
arguments of types that are promoted. [6471]
- Some compiler-generated wrapper functions with the
__STF____default_version
prefix were not resolved in template automatic instantiation mode. The
problem is corrected. [6362]
- Debugging support for constant variables is improved. Although the
compiler does not generate debug information for constant externs
(because you might not be able to link), it does generate debug
information for all constant variables whose underlying type is
int
or
float
- Bad code is no longer generated for an array reference within a
template instantiation. [6394]
- Runtime type identification has been corrected for shared images.
The RTTI calls
dynamic_cast
and
typeid
now work for objects created inside OpenVMS shared images.
With this correction, the name of virtual function tables have been
changed to
__vtbl2__<class>
, to prevent the new virtual function tables that point to RTTI
information from conflicting with old tables. [6348]
- Debugging support is now provided for anonymous union variables
inside namespaces. In the following example, debuggers support
referencing
s.a
and
s.b
:
struct S {
union {
int a;
int b;
}
} s;
|
In the following example, the compiler generates two variables,
x
and
y
, which the debuggers can examine.
union {
int x;
char y[4];
};
|
The compiler no longer generates tag names for tagless structs and
unions.
- Under the following conditions, the compiler could generate code
that returned the value of
i
before the store of the new value:
- Take the address of a variable, using a type other than the
variable's type,
(p = (char *)&i)
.
- Assign to the variable using a pointer addition expression,
*(p + 0) = new_value
.
- Fetch the variable's value directly
(return i)
.
The problem could occur only if the store using the pointer
addition expression appeared within an
if
statement, as in the following:
int f(int flag) {
int len = 1;
if (flag) {
char *ppp = (char *)&len;
*(ppp + 0) = 2;
};
return len;
}
|
The problem has been corrected. [6421]
- The compiler no longer generate incorrect code for offsets to
external arrays. [6386]
- The routine for
new[]
now calls
delete[]
if an exception occurs during construction. [6243]
- Many improvements have been made to cross-reference information
generated by the compiler and read by the SCA tool.
- The compiler now replaces embedded line feeds with blank spaces
during the initial scan.
- Cleaner header file inclusion policy
The new version of the
Standard Library is much more effecient than previous versions in its
header inclusion policy. For example, the header file
<algorithm>
no longer includes
<functional>
.
<ostream>
and
<istream>
no longer include
<locale>
. Programs that used to count on these inclusions might break. You can
correct them by explicitly including any header files you use in your
own sources.
- New interface to
get_temporary_buffer
and
use_facet
Because the Version 6.2 C++ compiler now supports explicit template
function arguments, it also supports the standard interface to the
get_temporary_buffer()
function. The following example shows how code must change: Change
this:
get_temporary_buffer(len,(T*)0); // two arguments
|
to this:
get_temporary_buffer<T>(len); // one argument
|
where "T" is the value type of the container. The standard
interface to the
locale
class
use_facet()
function is also now supported. The following example shows how code
must change: Change this:
use_facet(loc,(ctype<char>*)0);
|
to this:
// use_facet only takes one argument
use_facet<ctype<char> >(loc);
|
- Smaller executables for users of the
basic_string
library component
The Standard Library now provides a better
separation of the STL and Standard IO components. As a result, users of
the
basic_string
component of the Standard Library should obtain smaller executables. In
Version 6.1, using only a string in an application caused all
iostream
and
locale
object files from the Standard Library to be included in the
executable. These unnecessary files are no longer included. A
correction to the Standard Library helps programmers users of just the
basic_string
component to obtain smaller executables. In the previous version, using
just a string in an application caused the
iostream
and
locale
object files from the Standard Library to be included in the
executable. The new version corrects that problem.
-
money_get
/
money_put
locale
facets now conform to International C++ Standard
The
money_get
and
money_put
locale
facets have been corrected to match the standard. In the previous
version, for example,
money_get
appeared as:
template <class charT,
bool Intl = false,
class InputIterator = istreambuf_iterator<charT> >
class money_get;
|
They now correctly match the standard, where the interface appears
as:
template <class charT,
class InputIterator = istreambuf_iterator<charT> >
class money_get;
|
Note that the second template argument "Intl" has been removed. The
member functions
get()
and
put()
now accept
Intl
as an argument.
-
ios_base::openmode
flags set to conform to the International Standard
The Standard
Library file stream classes have been corrected to conform to the
International Standard with regard to setting the
ios_base::openmode
flags. In the previous release, it was possible to create a file for
reading and writing with this code:
#define __USE_STD_IOSTREAM
#include <stdlib.h>
#include <fstream>
int main() {
fstream fs("foo.out", ios_base::in | ios_base::out);
fs << "abc" << endl;
return EXIT_SUCCESS;
}
|
In the current release, this code works only if the file already
exists. If the file does not exist, you must also specify
ios_base::trunc
; that is, you must change the first line
in main()
to:
fstream fs("foo.out", ios_base::in | ios_base::out
| ios_base::trunc);
|
This conforms to table 92 in the Standard, which specifies the "C"
equivalent of the File open modes.
- Correction to
list::sort(Compare)
A bug in the
list::sort(Compare comp)
member function is corrected. Previously, if users supplied their own
Comparison
function object for the element of the list, the compiler issued a
message stating that it required an operator< defined for the
element type. This no longer occurs.
-
reverse_iterator
now matches the Standard
reverse_iterator
has been changed to match the standard. It now takes only one template
argument of type
iterator
instead of five. Users must change existing code to remove the
additional unnecessary arguments.
-
bitset
constructors no longer accept a
const char
A bitset can no longer be constructed with a
const char*
argument. For example, the following no longer compiles:
bitset<32> b("111111111");
|
The constructor that takes a string is a templatized constructor,
and thus can perform type deductions only on exact matches, not
conversions (for example,
const char*
to string). To make the code in the previous example compile with the
current version, the argument must be explicitly cast to a string, as
follows:
bitset<32> b( string("111111111"));
|
-
assign(size_t)
removed from
vector
,
deque
,
list
Previous releases of the Standard Library contained a member
function called assign() inside the
vector
,
deque
, and
list
classes. This function accepted only a
size_t
argument. This has been removed, because it is not in the Standard. You
must add an extra argument indicating the value you want assigned. For
example, you change calls like the following:
v.assign(5); // where v is a vector<int>
|
to:
-
allocator<>::deallocate(pointer)
removed
The member function
allocator<>::deallocate(pointer)
has been removed. The Standard requires two arguments for this member
function. The second argument should be of
size_type
and have the same value as the first argument passed to
allocator<>::allocate()
.
-
basic_ios
now initializes
skipws|dec
To conform to the Standard, the following
basic_ios
constructor constructs a
basic_ios
object and initializes the format control bits to
skipws | dec
:
explicit basic_ios(basic_streambuf<charT, traits>*sb)
|
Previously, this constructor also initialized the bit indicating
that output is right justified. Because the constructor is called while
constructing any of the IOStream objects
cout
,
clog
,
cerr
,
wcout
,
wclog
, or
wcerr
, the difference is apparent if you examine the format control bits set
after initializing one of these objects. Consider the following
program:
#include <stdlib.h>
#include <iostream>
using namespace std;
int main()
{
cout << cout.flags() << endl;
cout << clog.flags() << endl;
cout << cerr.flags() << endl;
cout << wcout.flags() << endl;
cout << wclog.flags() << endl;
cout << wcerr.flags() << endl;
return EXIT_SUCCESS;
}
|
The output now indicates that only the
skipws
and
dec
format control bits are initialized. Previously it would have indicated
that the right bit was also set.
- Some
iterator
classes removed
The following classes no longer exist in the
Standard and have been removed from library headers.
reverse_bidirectional_iterator
random_access_iterator
bidirectional_iterator
forward_iterator
output_iterator
input_iterator
Use instead the template class iterator with the template argument
category to indicate which type of iterator you are constructing.
- The default allocator argument changed for
basic_string
The default allocator argument for the class
basic_string
has been changed from
allocator<void>
to
allocator<charT>
. Any STL container constructed with an
allocator<void>
template argument no longer compiles, because the specialization of
allocator<void>
does not contain all the necessary typedefs.
-
strstream
now deletes underlying
strstreambuf
A problem has been corrected in the Standard Library
strstream
classes that prevented underlying
strstreambuf
(and thus the string) from being deleted when the
strstream
object was destroyed. The standard states that they should be deleted if
strmode & allocated
is true and
strmode & frozen
is not true. For example:
#define __USE_STD_IOSTREAM
#include <strstream>
void func()
{
ostrstream myostr;
myostr << "abc";
}
|
If you called
func()
the string "abc" was never deleted when the
myostr
stream was destroyed. This problem has been corrected. Note that the
Class Library
strstream
classes have always deleted the underlying string.
-
sync_with_stdio()
function is static
In previous versions, the function
sync_with_stdio()
was incorrectly declared as a member function of
ios_base
. The function is now correctly defined as a static member function; it
is no longer necessary to call it with the
"->"
or or
"."
notation.
10.3 Restrictions in Version 6.2
This section describes problems you might encounter when using the
current release of the C++ Standard Library with the HP C++
compiler. Where appropriate, workarounds are suggested.
- Do not use Standard Library template definition file names.
The C++ Standard Library supplies the following template definition files in SYS$LIBRARY:CXXL$ANSI_DEF.TLB:
algorithm.cc
|
fstream.cc
|
streambuf.cc
|
vector.cc
|
time.cc
|
bitset.cc
|
ios.cc
|
locimpl.cc
|
string.cc
|
ctype.cc
|
istream.cc
|
numbrw.cc
|
tree.cc
|
collate.cc
|
messages.cc
|
complex.cc
|
iterator.cc
|
ostream.cc
|
valarray.cc
|
money.cc
|
deque.cc
|
list.cc
|
sstream.cc
|
valimp.cc
|
numeral.cc
|
rwlocale.cc
|
|
|
|
|
If you use the same prefix name for any of your local files and have
the directory that contains them in your include search path, the automatic
instantiation mechanism picks up your local copy and does not find the library
files. (See
Compiling Programs with Automatic Instantiation
in
HP C++ User's Guide for OpenVMS Systems
.) It is best not to use any of these names as source file names
for your application.
- Redeclaration of Standard Library Functions
Many of the
prototypes in the Standard Library have been changed to conform to the
C++ International Standard by the addition of exception specifications.
This means that if you have redeclared the declarations in your own
code, you need to add the correct exception specification in order to
match what is declared in the header. For example:
#include <new.h>
// override default operator new
// this will give an error
inline void* operator new(size_t s);
|
To prevent this behavior, you must change your
new()
declaration to:
inline void* operator new(size_t s) throw(bad_alloc);
|
- Files/Macros for internal use only
HP C++ Version
6.n and higher ships the following non-Standard headers for HP
internal use only. Their contents are subject to change and can not be
relied on.
<stdcomp>, <stl_macros>, <stddefs>, <compnent.hxx>,
<stdmutex>, <stdexcept>, <lochelp>, <locimpl>, <locimpl.cc>,
<valimp>, <valimp.cc>, <vendor>, <codecvt>, <codecvt.cc>,
<collate>, <collate.cc>, <ctype>, <ctype.cc>, <locvector>,
<math>, <messages>, <messages.cc>, <money>, <money.cc>,
<numeral>, <numeral.cc>, <random>, <rwcats>, <rwlocale>,
<rwlocale.cc>, <rwstderr>, <rwstderr_macros>, <string_ref>,
<time>, <time.cc>, <traits>, <usefacet>
|
In addition both Standard and non-Standard headers make use of
macros beginning with
_RW
or
__RW
. These
_RW
* and
__RW
* macros are for HP internal use only. They are subject to change and
can not be relied on.
- Pre-ANSI iterators no longer available
The following classes
are no longer in the ANSI draft standard and should not be used:
input_iterator
output_iterator
forward_iterator
bidirectional_iterator
random_access_iterator
|
The functionality of these classes is now provided by the single
class iterator, which is templatized on the iterator category.
-
ctype_base::graph
In the current implementation of the
ctype_base
class, a character has a 'graph' property if and only if it also has an
'Alpha', a 'digit' or a 'punct' property. Thus, mathematical and
scientific symbols and dingbats from the UNICODE character set will not
be classified properly.
- IOStreams cannot output IEEE NaNs/Infinities
The Standard
IOStreams library does not support output for IEEE NaNs and Infinities.
-
ios_base::out
does not truncate a file to zero length
The
ios_base::openmode ios_base::out
should open a file for output. This means that the file is either
truncated to zero length if it exists or created for writing if it does
not. Therefore
ios_base::out
is the same as
ios_base::out
|
ios_base_trunc
. With our sources, the following program behaves incorrectly.
#include <stdlib.h>
#include <iostream>
#include <fstream>
using namespace std;
int main()
{
fstream fs ("t.in",ios_base::out);
fs << "A";
return EXIT_SUCCESS;
}
|
Where
t.in
contains xyz. After running this program,
t.in
contains
It should contain
You can work around this problem by replacing
ios_base::out
with
ios_base::out
|
ios_base::trunc
.
- Specifying a different
l_double_size
than the default size for your particular version of the operating
system does not work correctly with the standard library.
- Input and output to and from long double types on when specifying
/L_DOUBLE_SIZE=64
does not work if you use standard iostreams. This restriction will be
lifted in a future release.
- Overriding
operator new()
or
operator delete()
You can define a global
operator new()
or
operator delete()
to displace the versions used by the C++ Standard Library or C++ Class Library. For instructions, see
Overriding operator(new)
in
HP C++ User's Guide for OpenVMS Systems
.
- The
/template_define
and
/template_define=all
options are not guaranteed to work with the Standard Library. Specify
/template_define=used
instead.
When compiled with
/template_define
or
/template_define=all
, the following code generates compilation errors indicating that no
operator "<" (and ">" or "+" or "-") matches these operands:
#include <map>
map<int,int> foo;
|
The instantiation options are not guaranteed to work with the
Standard Library because they request the compiler to instantiate all
templates, even those that are not used. The
/template_define=all
option does not work because
rb_tree
, the underlying implementation of
map
and
set
supports a bidirectional iterator class. Thus,
operator+
,
operator-
,
operator<
and
operator>
are not defined in the iterator for that class. When you
instantiate the tree with
cxx /template_define=all
or
cxx /template_define
, the compiler attempts to instantiate recursively everything that is
typedef
ed, even if not used. Thus, the tree contains a
typedef
for
std::reverse_iterator<iterator>
, which then instantiates the global class
reverse_iterator
with the tree iterator as the template argument
RandomAccessIterator
, a misnomer in this case. This behavior generates undefined
symbols for these operators because they are used within the definition
of the operator member functions inside
reverse_iterator
. The compiler therefore attempts to instantiate them even though they
do not exist. Specifying
/template_define=used
for the Standard Library directs the compiler to only instantiate those
templates that are used.
- Description of the initial state of the
stringstream Ctor
There has been some controversy in the
comp.lang.c++
reflector and within the standards committee on the semantics of the
stringstream
constructor. Consider the following example:
#define __USE_STD_IOSTREAM
#include <stdlib.h>
#include <sstream>
int main() {
ostringstream ost("Hello, ");
// ost.rdbuf()->pubseekoff(0,ios::end,ios::out);
ost << "world!";
cout << ost.str() << endl;
return EXIT_SUCCESS;
}
|
Depending on the setting of the
streambuf put
pointer after the initial construction, the program could print either
"Hello, world!" or "world!". The Rogue Wave (and
HP C++) interpretation is that the
stringstream
constructor does not change the initial position of the
streambuf
pointer, so that the program prints "world!". If you want
to change the setting of the
put
pointer to match the other interpretation (that the pointer should move
to the end of the initializer string), you must insert a call to
pubseekoff()
, as shown in the commented line. If the ANSI C++ committee issues
a clarification on this matter HP C++ will implement their
decision.
-
basic_string
ambiguities
If you declare a
basic_string
of
int
you might encounter ambiguities in the constructors,
append
,
assign
,
insert
, and
replace
member functions. For example, if you write:
#include <stdlib.h>
#include <string>
int main() {
basic_string<int> si (5,0);
return EXIT_SUCCESS;
}
|
Compilation errors result from the overload resolution between
integral and iterator types. The constructors involved are the
following:
// construct n elements and initialize with value
basic_string(size_type n, charT c,
const Allocator& a = Allocator());
// construct using iterator ranges
template<class InputIterator>
basic_string(InputIterator begin, InputIterator end,
const Allocator& a = Allocator());
|
The compiler matches on the constructor
basic_string(InputIterator begin, InputIterator end, ...)
because
size_type
a
size_t
is an unsigned
int
on OpenVMS Alpha systems. So you have:
basic_string(unsigned int, int, ...) vs.
basic_string(int, int,...)
|
The second constructor is the better match, but it is undesirable
because we are not constructing via iterator ranges. The workaround
is to avoid matching on iterator types by casting integral arguments.
So for example, the previous program would compile correctly if the
size argument were cast to a
size_t
:
#include <stdlib.h>
#include <string>
int main() {
basic_string<int> si ((size_t)5,0);
return EXIT_SUCCESS;
}
|
- Restartable/non-restartable conversion routines in Locale classes
In
codecvt
class from localization library, the C++ standard assumes availability
of restartable conversion routines such as
mbrtowc
,
wcrtomb
, and so forth. These routines. introduced by ISO C Amendment 1: C
Integrity, were implemented in OpenVMS Alpha Version 6.2A and were never
ported back to the previous version. Because restartable conversion
routines are unavailable, the C++ Standard Library on OpenVMS
Version 6.2 uses their non-restartable counterparts such as
mbtowc
,
wctomb
and so forth. Note that restartable versions of conversion routines
are functional only when used with locales that support state-dependent
codeset encoding. Currently, neither the OpenVMS Alpha nor the Tru64
UNIX operating system provides any locales based on a
state-dependent codeset. With a codeset that is not
state-dependent, both restartable and non-restartable conversion
routines behave exactly the same. Therefore, using non-restartable
conversion routines instead of restartable ones does not constitute any
lack of functionality of C++ Standard Library on OpenVMS
Version 6.2.
-
cxxlink
errors with heavily templatized applications
When linking an
application that uses many templates, cxxlink might report and error
indicating that no I/O channels are available. For example:
$ cxxlink test
%LINK-F-OPENIN, error opening
T4:[TEST.CXX_REPOSITORY]CXX$CTQ173DGTWRTRB0K
64LIR.OBJ;1 as input
-RMS-F-CHN, assign channel system service request failed
-SYSTEM-F-NOIOCHAN, no I/O channel available
|
This error is generated when the process of linking all of an
application's template instantiations requires more open files than the
underlying system can support. In C++ Version 6.n and higher,
automatic template instantiation occurs at compile time (not link time
as in previous versions. Automatic template instantiation files are
therefore written into a repository as object files during compilation.
Cxxlink then invokes the OpenVMS linker to link each
instantiation file into the application. To work around this
problem, you must allocate sufficient system resources to handle an
extended number of open files. You might need to increase the SYSGEN
parameter CHANNELCNT, which specifies the number of permanent I/O
channels available to the system. You might also need to increase the
following process quotas:
open file quota (
fillm
)
buffered I/O quota (
biolm
)
page file quota (
pgflquo
)
enqueue quota (
enqlm
)
-
cxxlink
might fail with many large template instantiations
If you have many
large template instantiations in your source file, the cxxlink command
may fail with the following message:
%LINK-F-OPENIN, error opening <some repository> as input
-RMS-E-ACC, ACP file access failed
-SYSTEM-F-EXBYTLM, exceeded byte count quota
|
To work around this limit, you must increase your Buffered I/O byte
count quota, which you can see by entering the command SHOW
PROCESS/QUOTA. Ask your system administrator to use the Authorize
utility to increase the size. The recommended value is 199296.
- C++ Class Library dependencies on the Standard Library
As of
Version 6.0, the Class Library has dependencies on the Standard
Library. The dependencies have to do with incompatible name mangling
changes made between Version 5.n and Version 6.0 to address
Version 5.n name mangling problems. In all cases, entry points
to the Version 6.0 mangled names are available in the Version 6.0
Standard Library. This means that all Version 6.0 programs (even those
using only the Class Library) should be linked with the Standard
Library either with
cxxlink
or by adding
sys$library:libcxxstd.olb
to your link command. Using
cxxlink
is recommended. The following specific dependencies require support
from the Standard Library:
- Class library routines that use an ios::seek_dir type input
parameter:
istream::seekg()
ostream::seekp()
streambuf::seekoff()
filebuf::seekoff()
strstreambuf::seekoff()
stdiobuf::seekoff()
- In Version 5.n, the mangled name for an
enum
nested in a class did not contain the class name. This was corrected in
Version 6.0. Version 6.0 provides a switch (
/distinguish_nested_enums
) that allows you to compile programs amd generate either a Version
5.n or Version 6.0 style mangled name.
Because support for
the routines with Version 6.0 mangling resides in the Standard Library,
Version 6.0 programs making use of the above routines compiled using
/distinguish_nested_enums
must link with the Standard Library either with
cxxlink
or by inclusion of
sys$library:libcxxstd.olb
on the link line.
- Class Library inserter/extractor routines that accept
__int64
type parameters.
In Version 5.n, the mangled name for a
routine accepting an
__int64
parameter differed from the mangled name for a routine accepting a
long long
. This problem was corrected for Version 6.0 on OpenVMS.
Because support for the routines with Version 6.0 mangling resides in
the Standard Library, Version 6.0 programs making use of such routines
in code like the following
ifstream instream ("t.dat");
__int64 s_value
unsigned __int64 u_value
...
cout << u_value;
instream >> s_value;
|
must link with the Standard Library either with
cxxlink
or by inclusion of
sys$library:libcxxstd.olb
on the link line.
- Access violation when using Standard Library version of
cout
cin
, and other standard streams inside static constructors
If you are
using the standard iostreams library (if you have defined the macro
__USE_STD_IOSTREAM
) , and if you attempt to use
cout
,
cin
,
cerr
,
clog
,
wcout
,
wcin
,
wcerr
, or
wclog
in a static constructor, a core dump occurs. For example:
#include <ostream>
using namespace std;
struct C { C(); };
C::C() {
cout << "hello world"; // cout not initialized here
}
C c1;
int main() {}
|
To work around the problem, must modify your
cxxlink
step as follows:
- Create an options file,
image_name.opt
, whose content is
SYS$SHARE:LIBCXXSTD/INCLUDE=CXXL_STD_INIT
- Modify your cxxlink command to be:
$ CXXLINK/EXE=programname.exe image_name.opt/opt,[rest of command]
|
- Compilation warnings and errors with
auto_ptr
Using
auto_ptr
in non
strict_ansi
mode generates warnings about initializing a non-const ref with an
lvalue
. These warnings are due to the lack of enforcement of the rule in the
C++ standard that binding a reference to a non const to a class
rvalue
is illegal. To make
auto_ptr
work correctly, you must compile the module(s) that use
auto_ptr
s in
ansi
or
strict_ansi
language mode. In
strict_ansi
mode, you may still encounter compilation errors when converting an
auto_ptr<Derived>
to an
auto_ptr<Base>
. For example, this does not work:
struct Base {};
struct Derived : Base {};
auto_ptr<Derived> source2() {return auto_ptr<Derived>(new Derived);}
int main() {
auto_ptr<Derived> d;
auto_ptr<Base> b(d); // compiles
auto_ptr<Base> p3(source2()); // doesn't compile
return 0;
}
|
This is a known deficiency of the
auto_ptr
class, see language issue 84 at:
http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/core-issues.htm
|
for a discussion of the problem. You must either use casts or
avoid temporaries under these conditions. Remember that you should
not use
auto_ptr
s as elements of an STL container, because they do not meet the
CopyConstructible
and
Assignable
requirements for Standard Library container elements. Compilation
warnings occur if, for example, you try to insert an
auto_ptr
into a container. See
http://www.gotw.ca/gotw/025.htm
|
and
http://www.awl.com/cseng/titles/0-201-63371-X/auto_ptr.html
|
for discussions of the history and current restrictions of the
auto_ptr
class.
- The C++ International Standard permits overriding a virtual member
function based only on a derived class return type. The current release
does not support this capability.
- Instantiating function templates with array types can result in
different external name encoding than with C++ Version 5.n. To
avoid link errors, recompile the template definition with the current
version of the compiler.
- Specifying a different long double size than the default size for
your version of the operating system (using the
/l_double_size
qualifier) does not work correctly with the Standard Library.
11 Release Notes for the V6.0 C++ Compiler
11.1 Enhancements, Changes, and Restrictions in Version 6.0
This section briefly summarizes changes, enhancements, and restrictions
made in Version 6.0, including problems fixed.
- Support for the C++ International Standard (with some differences,
as described in Section 2.3), including the C++ Standard Library. See
Section 3 for information on and changes to the Standard Library.
- An updated debugger (dated 6-Jan-1999) that fixes problems with the
previous version of the debugger.
- Language mode options For compatibility with previous versions, the
compiler supports an ARM language mode and provides both
/standard=ansi
and
/standard=arm
language mode options, as well as options to support other C++ dialects. For details, see
Compiler Compatibility
in
HP C++ User's Guide for OpenVMS Systems
.
- Improved automatic instantiation of templates, including:
- Fewer restrictions. In particular, HP C++ no longer requires
that template declarations and definitions appear in header files.
- Build-time performance improvements that further reduce build times
for applications that use templates extensively.
For details, see
Using Templates
in
HP C++ User's Guide for OpenVMS Systems
.
- Language feature options
The compiler supports the following
options:
- Use of alternative tokens (see
Alternative Tokens
in
HP C++ User's Guide for OpenVMS Systems
).
- Run-Time type identification (see
Run-Time Type Identification
in
HP C++ User's Guide for OpenVMS Systems
).
- Restriction:
When using /EXTERN_MODEL=RELAXED_REFDEF (the
default) or /EXTERN_MODEL=COMMON, the C++ compiler describes each
global variable in the VMS Object language as using a unique Program
Section (PSECT). The name of the program section is the same as the
name of the variable, but it is subject to type encoding and
shortening. However, because the compiler must use several PSECTs with
fixed names to encode other aspects of the compiled program, these
fixed-name PSECTs are not available to users when these EXTERN_MODEL
values are in effect. All the names are prefixed and suffixed with a
dollar sign ( $ ) character, which generally identifies global names
reserved to OpenVMS system usage. The following fixed name PSECTs not
available as user variables with /EXTERN_MODEL=RELAXED_REFDEF or
/EXTERN_MODEL=COMMON: $ABS$, $BSS$, $CODE$, CXX$$RTTI, $DATA$,
$LINK$, $LITERAL$, $READONLY$, $TLS$
- Performance optimization options
Version 6.0 provides the
following performance optimization options:
- /[no]ansi_alias
- /assume=[no]pointers_to_globals
- /assume=[no]trusted_short_alignment
- /assume=[no]whole_program
For details, see
Performance Optimization Options
in
HP C++ User's Guide for OpenVMS Systems
.
The C++ Standard Library provided with this release defines a complete
specification (with some differences as described in Section 2.3) of
the C++ International Standard. The Standard Library in this release
includes for the first time the ANSI
locale
and
iostream
libraries.
Tutorial programs illustrating functionality found in the standard C++
library including the
locale
,
iostream
, and STL classes shipped with this release can be found in:
SYS$SYSROOT:[SYSHLP.EXAMPLES.CXX]*.CXX
|
You can compile and run these programs and use them as models for your
own coding. The expected output for each program can be found in:
SYS$SYSROOT:[SYSHLP.EXAMPLES.CXX]*.RES
|
Version 6.0 introduces the following major enhancements and changes. For detailed information about the HP C++ Standard Library, refer to
HP C++ User's Guide for OpenVMS Systems.
Additional changes include the following:
- Specific changes to match the C++ International Standard
-
iterator_traits::distance_type
now
iterator_traits::difference_type
The name of the
typedef
inside the classes
iterator_traits
and
iterator
that specifies the type of the result when two iterators are subtracted
has been changed from
distance_type
to
difference_type
.
-
typedef
name changes in
iterator
classes
The names of some of the typedefs in the
reverse_iterator
and
reverse_bidirectional_iterator
classes have changed as follows:
iter_type
is now
iterator_type
reference_type
is now
reference
pointer_type
is now
pointer
distance_type
is now
difference_type
-
slice/gslice
classes member function
length()
name change
The member function
length()
in the
slice
and
gslice
classes has changed its name to
size()
.
-
has_denorm
data member of the
numeric_limits
class changed
The
has_denorm
data member of the numeric_limits class,
<limits>
, has been updated.
has_denorm
is changed from type
bool
to an
enum
type
float_denorm_style
to reflect that support for denormalized values might not be detectable
at compile time. The
float_denorm_style
type looks like this:
namespace std {
enum float_denorm_style {
denorm_indeterminate = -1;
denorm_absent = 0;
denorm_present = 1;
};
}
|
The values representing the presence or absence of denorms are as
follows:
denorm_indeterminate
: cannot determine if type supports denormalized at compile time
denorm_absent
: the type does not support denormalized values
denorm_present
: the type supports denormalized values
- Default allocator value for
map
and
multimap
has changed
The default value for the template argument
Allocator
in map and multimap was changed from
allocator<T>
to
allocator<pair<const Key, T> >
.
- Interface Change for STL
distance()
function
Because the V6.0 compiler now supports partial
specialization of class templates, the interface to the
distance()
algorithm has been updated to conform to the latest C++ standard. This
means that previously, if you made a call to
distance
, the result was returned by using a reference argument for the third
argument:
// pre 6.0 the result was returned in d
distance(first,last,d);
|
Beginning with V6.0, the result is returned in the return type:
d = distance(first,last); // 6.0
|
You must therefore change all your calls to
distance()
.
- Support for
long long
and unsigned
long long
types.
The non-ANSI standard types
long long
and unsigned
long long
are now supported in the Standard Library iostreams as well as being
valid types for
numeric_limits
specializations and as types for which
destroy()
specializations are provided. For example, you can now say:
long long l;
cout << l << endl; // compiles without error
|
Note that these types are not supported in the
/standard=strict_ansi
compiler mode.
long long
and
unsigned long long
are also supported in the
iostream
class library in the default compiler mode.
- Common Instantiation Libraries no longer needed
Because Version
6.0 uses compile time rather than link time template instantiation,
there is no use for common instantiation libraries and we have
therefore not supplied any
build_common_instantiation_library
script. See the discussion of templates in
HP C++ User's Guide for OpenVMS Systems
.
|