C/C++ HDF5 Read string attribute

Question

A colleague of mine used labview to write an ASCII string as an attribute in an HDF5 file. I can see that the attribute exist, and read it, but I can't print it.

The attribute is, as shown in HDF Viewer:

Date = 2015\07\09

So "Date" is its name.

I'm trying to read the attribute with this code

hsize_t sz = H5Aget_storage_size(dateAttribHandler);
std::cout<<sz<<std::endl; //prints 16
hid_t atype = H5Aget_type(dateAttribHandler);
std::cout<<atype<<std::endl; //prints 50331867
std::cout<<H5Aread(dateAttribHandler,atype,(void*)date)<<std::endl; //prints 0
std::cout<<date<<std::endl; //prints messy characters!
//even with an std::string
std::string s(date);
std::cout<<s<<std::endl; //also prints a mess

Why is this happening? How can I get this string as a const char* or std::string?

I tried also using the type atype = H5Tcopy (H5T_C_S1);, and that didn't work too...

EDIT: Here I provide a full, self-contained program as it was requested:

#include <string>
#include <iostream>
#include <fstream>
#include <hdf5/serial/hdf5.h>
#include <hdf5/serial/hdf5_hl.h>

std::size_t GetFileSize(const std::string &filename)
{
    std::ifstream file(filename.c_str(), std::ios::binary | std::ios::ate);
    return file.tellg();
}

int ReadBinFileToString(const std::string &filename, std::string &data)
{
    std::fstream fileObject(filename.c_str(),std::ios::in | std::ios::binary);
    if(!fileObject.good())
    {
        return 1;
    }
    size_t filesize = GetFileSize(filename);
    data.resize(filesize);
    fileObject.read(&data.front(),filesize);
    fileObject.close();
    return 0;
}

int main(int argc, char *argv[])
{
    std::string filename("../Example.hdf5");
    std::string fileData;
    std::cout<<"Success read file into memory: "<<
               ReadBinFileToString(filename.c_str(),fileData)<<std::endl;

    hid_t handle;
    hid_t magFieldsDSHandle;
    hid_t dateAttribHandler;
    htri_t dateAtribExists;

    handle = H5LTopen_file_image((void*)fileData.c_str(),fileData.size(),H5LT_FILE_IMAGE_DONT_COPY | H5LT_FILE_IMAGE_DONT_RELEASE);
    magFieldsDSHandle = H5Dopen(handle,"MagneticFields",H5P_DEFAULT);
    dateAtribExists = H5Aexists(magFieldsDSHandle,"Date");
    if(dateAtribExists)
    {
        dateAttribHandler = H5Aopen(magFieldsDSHandle,"Date",H5P_DEFAULT);
    }


    std::cout<<"Reading file done."<<std::endl;
    std::cout<<"Open handler: "<<handle<<std::endl;
    std::cout<<"DS handler: "<<magFieldsDSHandle<<std::endl;
    std::cout<<"Attributes exists: "<<dateAtribExists<<std::endl;
    hsize_t sz = H5Aget_storage_size(dateAttribHandler);
    std::cout<<sz<<std::endl;
    char* date = new char[sz+1];
    std::cout<<"mem bef: "<<date<<std::endl;
    hid_t atype = H5Aget_type(dateAttribHandler);
    std::cout<<atype<<std::endl;
    std::cout<<H5Aread(dateAttribHandler,atype,(void*)date)<<std::endl;
    fprintf(stderr, "Attribute string read was '%s'\n", date);
    date[sz] = '\0';
    std::string s(date);
    std::cout<<"mem aft: "<<date<<std::endl;
    std::cout<<s<<std::endl;

    H5Dclose(magFieldsDSHandle);
    H5Fclose(handle);


    return 0;
}

Printed output of this program:

Success read file into memory: 0
Reading file done.
Open handler: 16777216
DS handler: 83886080
Attributes exists: 1
16
mem bef: 
50331867
0
Attribute string read was '�P7'
mem aft: �P7
�P7
Press <RETURN> to close this window...

Thanks.

Answer 1

It turned out that H5Aread has to be called with a reference of the char pointer... so pointer of a pointer:

H5Aread(dateAttribHandler,atype,&date);

Keep in mind that one doesn't have to reserve memory for that. The library will reserve memory, and then you can free it with H5free_memory(date).

This worked fine.

EDIT:

I learned that this is the case only when the string to be read has variable length. If the string has a fixed length, then one has to manually reserve memory with size length+1 and even manually set the last char to null (to get a null-terminated string. There is a function in the hdf5 library that checks whether a string is fixed in length.

Answer 2

I discovered that if you do not allocate date and pass the &date to H5Aread, then it works. (I use the C++ and python APIs, so I do not know the C api very well.) Specifically change:

char* date = 0;
// std::cout<<"mem bef: "<<date<<std::endl;    

std::cout << H5Aread(dateAttribHandler, atype, &date) << std::endl;

And you should see 2015\07\09 printed.

You may want to consider using the C++ API. Using the C++ API, your example becomes:

std::string filename("c:/temp/Example.hdf5");
H5::H5File file(filename, H5F_ACC_RDONLY);
H5::DataSet ds_mag = file.openDataSet("MagneticFields");

if (ds_mag.attrExists("Date"))
{
    H5::Attribute attr_date = ds_mag.openAttribute("Date");
    H5::StrType stype = attr_date.getStrType();
    std::string date_str;
    attr_date.read(stype, date_str);
    std::cout << "date_str= <" << date_str << ">" << std::endl;
}

Answer 3

As a simpler alternative to existing APIs, your use-case could be solved as follows in C using HDFql:

// declare variable 'value'
char *value;

// register variable 'value' for subsequent use (by HDFql) 
hdfql_variable_register(&value);

// read 'Date' (from 'MagneticFields') and populate variable 'value' with it
hdfql_execute("SELECT FROM Example.hdf5 MagneticFields/Date INTO MEMORY 0");

// display value stored in variable 'value'
printf("Date=%s\n", value);

FYI, besides C, the code above can be used in C++, Python, Java, C#, Fortran or R with minimal changes.