haskey edge cases

[simonbyrne]

It looks like haskey(::Union{Group,File}, ::String) has some edge cases that aren't currently well-handled:

using HDF5

filename = tempname()
f = h5open(filename, "w")
f["aa"] = 1
f["bb/cc"] = 2

# the following all throw errors
haskey(f, "./aa")
haskey(f, "aa//")
haskey(f, "aa/c")
haskey(f, "bb//cc")

There is some description of how it should be used here: https://portal.hdfgroup.org/display/HDF5/H5L_EXISTS, and the h5py code which does this is here:
https://github.com/h5py/h5py/blob/6ffb48b5ccb5f14f4fe1208b27e64aae9c85fd48/h5py/h5g.pyx#L495

[mkitti]

julia> haskey(f, "./aa")
false

This does not throw an error, but perhaps it should be true?

julia> HDF5.API.h5l_exists(f, "./aa", HDF5.API.H5P_DEFAULT)
true

julia> HDF5.API.h5l_exists(f, "aa//", HDF5.API.H5P_DEFAULT)
true

julia> HDF5.API.h5l_exists(f, "bb//cc", HDF5.API.H5P_DEFAULT)
true

julia> HDF5.API.h5l_exists(f, "aa/c", HDF5.API.H5P_DEFAULT)
ERROR: HDF5.API.H5Error: Cannot determine whether aa/c exists
...

Should "aa/c" not error? It does not exist.

[simonbyrne]

Should "aa/c" not error? It does not exist.

I would say it should return false.

[mkitti]

In #978 and #979, I'm expanding and updating the low level H5O bindings, which seem relevant here based on the h5py code.

[mkitti]

At the end of the loop, it looks like we will need API.h5o_get_info_by_name to check if the object is a group before iterating again.

We will need to write a high level hdf5type function that abstracts over the different h5o_get_info_by_name versions (1 vs 3). For version 3, we can use H5O_INFO_BASIC as the fields argument.

[mkitti]

Here's a link to has_key in this code base. The function is recursive rather than iterative.