Variable length string Read not working

[tclarke]

Read() of scalar variable length string data does not appear to work.

var buf string
dset.Read(buf, hdf5.T_GO_STRING)

will error with:
panic: reflect.Value.UnsafeAddr of unaddressable value [recovered]
panic: reflect.Value.UnsafeAddr of unaddressable value

using dset.Read(&buf, hdf5.T_GO_STRING) works with uint, etc. but not with string. There is not error but buf will be empty.

[naegelejd]

I can reproduce this issue. I create a datset with a simple 1-D dataspace, with datatype hdf5.T_GO_STRING, then write a string slice containing only my single string to the dataset (errors ignored for simplicity):

dataspace, _ := hdf5.CreateSimpleDataspace([]uint{1}, nil)
dataset, _ := file.CreateDataset("group/xml", hdf5.T_GO_STRING, dataspace)
wrapper := []string{xml}
dataset.Write(&wrapper)

This works great! However, when I want to read my string back from the dataset I get an empty string:

dataset, _ := file.OpenDataset("group/xml")
wrapper := make([]string, 1)
dataset.Read(&wrapper)
fmt.Println(wrapper[0])

[naegelejd]

In fact, the strings in the s1Type in your example here are not properly read back into memory after being written to the HDF5 file. If you check the output printed here, you'll see that each struct's string member e is "empty". Also, each array member's last item is not read properly (it remains the zero value). You can see the difference between what is written and what is read in the output of your example:

:: data: [{0 0 1 [0 0 0] --0--} {1 1 0.5 [1 2 3] --1--} {2 4 0.3333333333333333 [2 4 6] --2--} {3 9 0.25 [3 6 9] --3--} {4 16 0.2 [4 8 12] --4--} {5 25 0.16666666666666666 [5 10 15] --5--} {6 36 0.14285714285714285 [6 12 18] --6--} {7 49 0.125 [7 14 21] --7--} {8 64 0.1111111111111111 [8 16 24] --8--} {9 81 0.1 [9 18 27] --9--}]
:: file [SDScompound.h5] created (id=16777216)
:: dset (id=83886080)
:: dset.Write...
:: dset.Write... [ok]
:: data: [{0 0 1 [0 0 0] } {1 1 0.5 [1 2 0] } {2 4 0.3333333333333333 [2 4 0] } {3 9 0.25 [3 6 0] } {4 16 0.2 [4 8 0] } {5 25 0.16666666666666666 [5 10 0] } {6 36 0.14285714285714285 [6 12 0] } {7 49 0.125 [7 14 0] } {8 64 0.1111111111111111 [8 16 0] } {9 81 0.1 [9 18 0] }]

I'm not positive, but, at least for strings and string slices, I don't think you can just call H5Dread to fill a string's StringHeader.Data field without updating its StringHeader.Len field. If I manually update this field in your Dataset.Read after the call to H5Dread, the string looks valid... but this is super hacky and just to demonstrate my point:

diff --git a/h5d.go b/h5d.go
index a3fa9d9..f22059e 100644
--- a/h5d.go
+++ b/h5d.go
@@ -149,6 +149,20 @@ func (s *Dataset) WriteSubset(data interface{}, memspace, filespace *Dataspace)
                filespace_id = filespace.id
        }
        rc := C.H5Dwrite(s.id, dtype.id, memspace_id, filespace_id, 0, addr)
+
+       // only checking for string slices
+       sp, ok := data.(*[]string)
+       if ok {
+               // show that the string was actually read from the file
+               fmt.Print("C.puts(...): ")
+               C.puts(*(**C.char)(addr))
+
+               // make Go believe it's a string of len > 0
+               head := (*reflect.SliceHeader)(unsafe.Pointer(&(*sp)[0]))
+               head.Len = int(C.strlen((*C.char)(unsafe.Pointer(head.Data))))
+               fmt.Println("s[0]: ", (*sp)[0])
+       }
+
        err = h5err(rc)
        return err
 }

This doesn't work, for example, with a struct containing strings, as in your example