Description:
Merged from trunk r28027.
- Removed H5Library::instance because it is unnecessary. All H5Library's
methods are static. This, in turn, removed the memory leaks by
H5Library::instance not being deleted.
- Added ObjCreatPropList::deleteConstants to atexist() list
- Cleaned up comments and format inconsistencies with 1.8
Platforms tested:
Linux/32 2.6 (jam)
Linux/64 (platypus)
Darwin (osx1010test)
Merged from trunk r27945, which was merged from hdf5_CppAPI_Constants,
including the revisions listed below.
Platforms tested:
Linux/32 2.6 (jam)
Linux/64 (platypus)
Darwin (osx1010test)
+++ Log message when merged from hdf5_CppAPI_Constants to trunk, r27945 +++
Purpose: Fix HDFFV-9529
Description:
Merged from hdf5_CppAPI_Constants
r27942:
------
Description:
- Added H5dont_atexit() to getPredType and all the getConstant's to prevent
the C library from terminating before the C++ library cleanup.
- More cleanup and added more comments
k
r27923:
------
- Updated more comments and moved some things around for consistency
- Removed check for "new" failure, exceptions would be thrown
r27922:
------
Description:
Added function headers and more comments for clarity.
r27917:
------
Description:
The C++ library has several types of global constants from different
classes, such as PropList, PredType, DataSpace, etc... Previously,
these global constants were declared statically and the C++ library used
a constant, called PredType::AtExit, to detect when all the global
contants are destroyed then close the C library (H5close). This method
relied on the order of the constants being created and destroyed and
that PredType constants be the last to be destroyed. In September 2015,
it was recognized that the order in which the global constants were
created and destroyed was actually undefined, thus can be different
between different compilers. This resulted in failure when compilers
destroy PredType constants before others because when PredType::AtExit
was destroyed, the C library was closed, so when the constants of other
classes such as PropList or DataSpace were being deleted, the C library
would not be available.
Solution:
The static approach is changed to dynamic. In order to avoid an impact
on existing applications, the static global constants are changed to
constant references to the dynamically allocated objects.
A detailed explanation of the new method and a description of the
changes are in a Design Notes at the end of the file H5PredType.cpp.
New functions added to support the new methods are listed below.
class H5Library:
// Returns a singleton H5Library to initialize the global
// constants, invoked in IdComponent default constructor
static H5Library* getInstance(); // public
// Registers cleanup and terminating functions with atexit(),
// called in IdComponent default constructor
static void initH5cpp(void); // public
// Calls H5close to terminate the library, registered with
// atexit(), as the last thing to be done.
static void termH5cpp(void); // public
class PredType:
// Creates the constants
static void makePredTypes(); // private
// Calls makePredTypes to create the constants and returns
// the dummy constant PREDTYPE_CONST;
static PredType* getPredTypes(); // private
class DataSpace:
// Creates the constant
static DataSpace* getConstant(); // private
class PropList:
// Creates the constant
static PropList* getConstant(); // private
class DSetCreatPropList:
// Creates the constant
static DSetCreatPropList* getConstant(); // private
class DSetMemXferPropList:
// Creates the constant
static DSetMemXferPropList* getConstant(); // private
class FileCreatPropList:
// Creates the constant
static FileCreatPropList* getConstant(); // private
class FileAccPropList:
// Creates the constant
static FileAccPropList* getConstant(); // private
This function is added to PredType, DataSpace, PropList, and the four
subclasses of PropList:
// Deletes the constant
static void deleteConstants(); // public
Description:
When copy constructor or constructor that takes an existing id is invoked,
the C ref counter stays the same but there is an extra C++ object which
later is destroyed and may cause the HDF5 id to be closed prematurely. The
C++ library needs to increment the ref counter in these situations, so that
the C library will not close the id when it is still being referenced.
However, the incrementing of ref count left some objects opened at the end
of the program, perhaps, due to compiler's optimization on cons/destructors. The constructor, that takes an existing id, needs to increment the counter
but it seems that the matching destructor wasn't invoked. The workaround
is to have a function for each class that has "id" that only sets the id
and not increment the ref count for the library to use in these situations.
These functions are "friend" and not public.
The friend functions are:
void f_Attribute_setId(Attribute *, hid_t)
void f_DataSet_setId(DataSet *, hid_t)
void f_DataSpace_setId(DataSpace *, hid_t)
void f_DataType_setId(DataType *, hid_t)
Merged from trunk: r26655
Platforms tested:
Linux/64 (platypus)
Linux/32 2.6 (jam Intel 15.0)
SunOS 5.11 (emu)
Description: (r26457)
The following two constructors of classes: AbstractDs, IdComponent,
H5Location, and H5Object are no longer appropriate after the data member
"id" had been moved from IdComponent to the sub-classes.
The copy constructors are noop and will be generated by the compilers if
needed so they can be removed in 1.8.15 without effecting applications. The
other constructors will be removed from 1.10 release, and then from 1.8 if
their removal does not raise any problems in at least two 1.10 releases.
<Classname>(const hid_t h5_id);
<Classname>(const <Classname>& original);
Deprecation warnings are added to each of these classes.
The other classes are changed to call the default constructor instead of
one of the constructors above.
Platforms tested:
Linux/64 (platypus)
Linux/32 2.6 (jam)
SunOS 5.11 (emu)
- Added wrappers to H5Object for H5Iget_name() to get object's name
ssize_t getObjName(char *obj_name, size_t buf_size = 0) const;
ssize_t getObjName(H5std_string& obj_name, size_t len = 0) const;
H5std_string getObjName() const;
- Added tests tobject.cpp
- Added to various cleanup_* functions in tests to remove generated files
- Added an overload H5I_type_t getHDFObjType() to get object's type
(merge from trunk-24969)
Platforms tested:
Linux/ppc64 (ostrich)
Linux/32 2.6 (jam)
SunOS 5.11 (emu) with gmake
- Added a lot of documentation to classes for Reference Manual.
- Fixed some format inconsistencies
Platforms tested:
Linux/ppc64 (ostrich)
Linux/32 2.6 (jam)
Linux/64 2.6 (koala)/PGI compilers
Description from trunk, r22836
In this bug, H5File doesn't have the ability to create attribute. The
following changes will provide that functionality and several others that
were also missing:
- Added an abstract class H5Location in between IdComponent and H5Object.
- New class structure of IdComponent, H5Location, H5Object, H5File
IdComponent
|
H5Location
/ \
H5Object H5File
- Wrappers in H5Object were moved to H5Location because the related C
functions take either file, group, dataset, or named datatype ID.
- Added wrapper for H5Rget_obj_type2
- Added tests for file attributes and H5Rget_obj_type2 wrapper
Description from trunk, r22845
Fixed miscellaneous inconsistencies and typos, which also took
care of the failure in Packet Table test on daily test today.
Description from trunk, r24143
Fixed comments, documentation, and mis-matched DOXYGEN_SHOULD_SKIP_THIS pairs.
Description from trunk, r24188
- The failure in daily test was caused by missing initialization of member
"id" in a few constructors. This is now fixed.
- Added two overloaded H5Location::setComment
- Improved some error reporting in H5Location
- Improved error reporting in tests
Description from trunk, r24189
Changed header guards from single underscore to double underscore.
Platforms tested:
SunOS 5.11 (emu)
Linux/32 2.6 (jam)
Linux/64 2.6 (koala)/PGI compilers
Oops, I accidentally only did "svn commit" instead of "svn commit -m" as
instructed.
"ported revisions 15958 to 15959 (Added missing wrappers for H5Rdereference)
from trunk."
Description:
Accidentally left p_setId in public section in the previous checkin.
Moved it to "protected:" section.
Platforms tested:
Linux 2.6 (kagiso)
SunOS 5.10 (linew)
FreeBSD (duty)
Description:
Changed all subclasses' setId to p_setId and put back setId in
IdComponent. p_setId is used in the library where the id provided
by a C API passed on to user's application in the form of a C++ API
object, which will be destroyed properly, and so p_setId does not
call incRefCount. On the other hand, the public version setId is
used by other applications, in which the id passed to setId needs
to be closed properly by the application, so setId must call incRefCount
for the new object to prevent prematurely closing of the id.
Platforms tested:
Linux 2.6 (kagiso)
Still on going with these two platforms, but needed to check in before
daily tests start.
SunOS 5.10 (linew)
FreeBSD (duty)
Description:
The class hierarchy was revised to address the problem reported in
bugzilla #1068. Classes AbstractDS and Attribute are moved out of
H5Object. Class Attribute now multiply inherits from IdComponent and
AbstractDs and class DataSet from H5Object and AbstractDs.
In addition, data member IdComponent::id was moved into subclasses:
Attribute, DataSet, DataSpace, DataType, H5File, Group, and PropList.
Also fixed bugzilla 1045: revised Attribute::write and Attribute::read
wrappers to handle memory allocation/deallocation properly. (bugzilla 1045)
Platforms tested:
SunOS 5.10 (linew)
Linux 2.6 (kagiso)
FreeBSD (duty)
Add H5Rget_obj_type() to the API versioning and switch internal
routines to use H5Rget_obj_type2()
Misc. other code cleanups, etc.
Tested on:
FreeBSD/32 6.2 (duty)
FreeBSD/64 6.2 (liberty)
Linux/32 2.6 (kagiso)
Linux/64 2.6 (smirom)
AIX/32 5.3 (copper)
Solaris/32 2.10 (linew)
Mac OS X/32 10.4.10 (amazon)
Tested platform:
Kagiso only since it is only a comment block change. If it works in one
machine, it should work in all, I hope. Still need to check the parallel
build on copper.
Description:
Fixed typos and re-arranged some functions to be in sync with 1.6.
Platforms tested:
Linux 2.4 (heping)
SunOS 5.9 (shanti)
AIX 5.1 (copper)
Description:
Changed to alias string instead of std, i.e. H5std_string instead
of H5std, because the old way wasn't working when std didn't exist.
Platforms tested:
Linux 2.4 (heping)
SunOS 5.8 64-bit (sol)
HPUX 11.00 (kelgia) - this was the problematic platform but I wasn't
able to test before.
Description:
Added alias H5_std so either the global or std namespace can be
used, depending on H5_NO_STD.
Platforms tested:
Linux 2.4 (heping)
SunOS 5.8 64-bit (sol)
AIX 5.1 (copper)
Description:
There was a workaround for predefined types, in the C++ library,
implemented when the C++ library was handling the reference
counting of the object ids on its own. Currently, the C++ library
is using the available APIs from the C library for that purpose,
and there were bugs reported involving that part of the C++ library.
So, I decided to remove the workaround completely.
Also, improved the use of std members.
Platforms tested:
Linux 2.4 (heping)
SunOS 5.8 64-bit (sol)
Linux 2.4 w/PGI (colonelk)
Description:
Added wrapper for H5Iget_type.
Added try/catch to many APIs that call private functions so that more
specific information can be provided at failure.
Added IdComponent::inMemFunc to help providing specific info.
Added const to parameters of several functions that missed that.
Platforms tested:
Linux 2.4 (heping)
SunOS 5.8 64-bit (sol)
AIX 5.1 (copper)
IRIX64 with -n32 (modi4)
HPUX 11.00 (kelgia)
Description:
The use of FileCreatPropList::DEFAULT sometimes caused failure
in the reference counting area. This occurs to all the default
property lists, which also include FileAccPropList::DEFAULT,
DSetCreatPropList::DEFAULT, and DSetMemXferPropList::DEFAULT.
H5P_DEFAULT was used to create these default prop lists and
because its value is 0, the id of these prop lists are 0, which
is rejected by the H5I functions during the reference counting.
Solution:
The main action to fix the above problem was to use
H5P_FILE_CREATE
H5P_FILE_ACCESS
H5P_DATASET_CREATE
H5P_DATASET_XFER
to define the default property lists accordingly. Yet, when this
fix was applied, some bug in reference counting was revealed.
It appeared that some ids were not incremented but were passed in
for decrementing. The following actions were then taken to fix and
improve the current use of reference counting H5I functions.
* added private func IdComponent::p_valid_id to verify that the
id is a valid id and can be passed into an H5I function
* used p_valid_id to validate an id before calling an H5I functions
in the reference-counting member functions incRefCount,
decRefCount, and getCounter
* changed to use member function incRefCount, decRefCount, and
getCounter instead of the C APIs H5Iinc_ref, H5Idec_ref, and
H5Iget_ref throughout IdComponent.
In addition, overloadings were added for incRefCount, decRefCount,
and getCounter to take an id different than the id of the current
instance; they can be convenient during debugging.
Platforms tested:
Linux 2.4 (heping)
SunOS 5.8 64-bit (sol)
AIX 5.1 (copper)
Description:
Added wrappers for H5Fget_name:
H5File::getFileName
H5Object::getFileName
Moved some functions from Group into the base class CommonFG for
H5File too.
Platforms tested:
SunOS 5.7 (arabica)
Linux 2.4 (eirene)
Code update
Description:
Update C++ wrappers to use new ID reference counting mechanisms.
Platforms tested:
FreeBSD 4.9 (sleipnir) w/C++
Too focused to require h5committest
Add/Improve documentation
Description
Added doxygen documentation to H5PredType.cpp and made minor changes
to the others for either correction or clarification.
Platforms:
SunOS 5.7 (arabica)
Linux 2.4 (eirene)
Misc. update:
Add C++ wrappers - incrementally check-in
Description
Added wrapper for these C APIs:
H5Rcreate
H5Rget_obj_type
H5Rget_region
This is an incremental check-in to preserve the code, corresponding
tests will follow in a few weeks.
Platforms:
Linux 2.4 (eirene)
I'm checking the code out and test on arabica too.
Misc. update:
Purpose:
use H5_DLLCPP to replacer __DLLCPP__ for c++ interfaces.
Description:
Solution:
Platforms tested:
linux 2.2.18smp, IRIX64, solaris 2.7, windows 2000
Purpose:
Adding support for dll
Description:
Added __DLLCPP__ to all public classes and templates.
Added #include "H5Include.h" to H5RefCounter.cpp because of the
use of DLLCPP in .h file and it needs the following chain:
H5Include.h/hdf5.h/H5public.h/H5api_adpt.h
Added #pragma warning(disable: 4251) to H5Exception.h to eliminate
this warning on private data members of type 'string.' This
occurs because 'string' is not yet instantiated at compilation
time; however, since the class is exported, the warning is
harmless.
Changed this member function's parameter to be passed as reference.
from:
void CompType::insertMember( const string name,...
to:
void CompType::insertMember( const string& name,...
Platforms tested:
Linux 6.2 (eirene)
Windows 2000
Purpose:
Bug fixing
Description:
H5IdComponent::operator= shouldn't be virtual because the subclass'
operator= have different type for the rhs argument. Removed virtual
Platforms tested:
SunOS 5.7 (arabica)
IRIX64 (modi4) - only tested that the warnings went away, still working
on other configuration problems.
Description and Solution:
- Predefined type objects become outdated when the application calls
H5close and then H5open to use the library again.
Solution:
+ Reimplemented PredType using enum type to obtain up-to-date
values for predefined types, i.e. H5T_xxx_yyy's, when they
are used.
+ Added to PredType its own version of getId() to return the
appropriate predefined type id according to the enum value.
+ Made IdComponent::getId() a virtual function
- Since Albert already fixed the "multiply defined RcsId" problem,
removed the now become unnessary "#undef RCSID" from H5Include.h
- Added a check for func_name != NULL in one of the Exception
constructors because seg. fault. might occur in some situations.
- It is more convenient to make a copy of a datatype, dataspace,
or property list by using assignment operator
Solution:
+ Added assignment operator to DataType, DataSpace, and
PropList so that they will invoke H5Tcopy, H5Scopy,
and H5Pcopy, respectively, instead of just simply
copying the data members as for other classes. So,
now a type, space, or property list can be copied by
either :
mytype = original_type, or
mytype.copy(original_type)
Note that copy constructors are still just copying the
data members.
+ Added an assignment operator to DataType that takes a
predefined type.
+ Made IdComponent::operator= a virtual function
Platforms tested:
arabica (sparc-sun-solaris 2.7)
Feasibility improvement
Description:
- Some member functions in class Exception can be called without
an Exception instance exits, but because they are not static, they
cannot.
- Many exception throwings don't provide any information to callers.
Solution:
- Add 'static' to several member functions in class Exception.
- Added <class name::function name> to some exceptions thrown
in H5File.cpp and H5FcreatProp.cpp. I'm trying this for users'
comments. More information may be added later and to all
exceptions.
- Changed the comment line for emacs editor to:
// C++ informative line for the emacs editor: -*- C++ -*-
because... I'd rather not say that the code "may look like C" :-)
Platforms tested:
Linux (gcc version egcs-2.91.66)
Bug Fix
Description:
Added comments to the top of C++ files to indicate that it's a C++
file and not a C source code file. Also added a macro which makes
emacs automagically enter c++-mode. (Recommended by Frank Schimmel
fschimme@monsoon.CAPS.ou.edu)
Fix and improve
Description:
- Put functions that are common to H5File and Group into a
prototype class, CommonFG. I didn't do that before because
of the fear of the consequences of multiple inheritance, since
H5File and Group already inherit from different super classes.
I recently read a C++ book and learned to use MI more safely.
This change reduced some more of code redundancy.
- Added missing const to some function parameters
Platforms tested:
Solaris/CC 5.0 (arabica)
Description:
Dec CC compiler doesn't support some new c++ features.
Some typos caused compilation erros on Dec Alpha.
Solution:
- Added macro BOOL_NOTDEFINED to define bool type on Dec
Alpha (gondolin) since its compiler doesn't support bool.
- Added macro NO_STATIC_CAST to prevent the use of the new
c++ feature static_cast since Dec CC compiler doesn't support.
- Added const to parameters of some functions to match the
functions' declaration and definition. Typos errors and
Dec CC complained.
Platform:
Solaris (arabica). Also tried to build on Dec Alpha but still
not linked due to some missing compiler flags; its compilation
went fine though.
C++ API for 1.3.x branch
Description:
The *.C and *.h files named different than those in 1.2.x.
They are in the form: 'H5' + classname, or just classname if
the classname is already prefixed with 'H5' to avoid ambiguity
in documentation context. This version has several hidden bugs
fixed and an improvement on the reference counting approach.
The classes and their inheritance structure are listed below:
---------------------------------------
H5Library
Exception
RefCounter
IdComponent
H5File
DataSpace
H5Object
Group
AbstractDs
DataSet
Attribute
DataType
PredType
EnumType
CompType
AtomType
StrType
IntType
FloatType
PropList
FileCreatPropList
FileAccPropList
DSetCreatPropList
DSetMemXferPropList
---------------------------------------
IdComponent uses RefCounter to keep track of opened objects
so proper termination of HDF5 objects can be maintained.
Each class has a .h file containing the class declaration and
a .C file containing its definition. In addition to the classes
files, the following files do not have class information:
- H5Cpp.h: header file to be included in user's application
- H5Idtemplates.h: contains a template function used by several classes
- H5Classes.h: contains forward class declarations
- H5CommonFG.*: contains common code used by classes H5File and Group
- H5Include.h: contains the hdf5.h header file and the #undef RCSID
to work around the problem: multiple defined RcsId
- H5Alltypes.h: simply serves as a container to hold the header
files of all datatypes to simplify the header file inclusion
Platforms:
Solaris (arabica) and Linux
