Enable empty string attributes

[franzpoeschel]

Close #1329.
I think the original issue why we had this restriction in the first place can be understood from the documentation of H5Tset_size in HDF5:

size must have a positive value, unless it is passed in as H5T_VARIABLE and the datatype is a string datatype.
[…]
String or character datatypes: The size set for a string datatype should include space for the null-terminator character, otherwise it will not be stored on (or retrieved from) disk. Adjusting the size of a string automatically sets the precision to 8*size.

• HDF5 supports storing null-terminated strings as well as non-null-terminated strings, but recommends using null-terminated strings
• Non-variable strings (which are used by openPMD) must have a positive size >0, meaning that empty strings must be null-terminated
• openPMD so far used non-null-terminated strings, so storing empty strings was not possible

This commit switches to using null-terminated strings.

• Merge Parsing logic: fail gracefully on unexpected input #1237 first

[franzpoeschel]

Subtle bug fixed in HDF5IOHandler writeAttribute() method:
new char[…] does not initialize the memory, but strncopy will copy until either a NULL is found or the max_len has been reached, so the uninitialized memory is always guarded from actually being read into a std::string. Also, no segfault is triggered, since the memory left and right of the uninitialized memory is written and the virtual memory can detect no illegal access in that case.

Before this PR, a std::vector{"VECTOR", "OF", "STRINGS"} was converted to the following c_str buffer:

VECTOR0
OF0****
STRINGS

where 0 is a null terminator and * is uninitialized memory. The null terminator is missing for the longest string, all other strings have one null terminator, followed by uninitialized memory.

With this PR, it is converted to:

VECTOR00
OF000000
STRINGS0

Every line has at least one null terminator, and the rest of the line is also padded with nulls.

[ax3l]

Awesome, thanks a lot for the fix and explanation.

[franzpoeschel]

Hmm, ADIOS1 seems to not like it. We are deprecating that anyway, but may I can find a solution even.

[ax3l]

Awesome, thanks a lot!