Fix for reading zero-length values using RawRead

[missinglink]

When reading zero-length values using RawRead (ie. nil, empty) the getBytes() function triggers a fatal SIGBUS error.
We can avoid this by detecting the case where val.mv_size == 0 and avoiding attempts to access that memory.

resolves #25
ref: https://github.com/ledgerwatch/lmdb-go/blob/master/lmdb/lmdb.h#L255

[AskAlexSharov]

@missinglink - how to separate Nil from empty []byte{} ? Does MDBX support both?

[missinglink]

LMDB doesn't seem to have a distinction between nil and empty 🤷

[missinglink]

The C value val.mv_size is equal to 0 for values stored as either var v_nil []byte or var v_empty = []byte{}, it doesn't provide any way of distinguishing between the two.

[missinglink]

But reading either form prior to this PR (in combination with RawRead) triggers a panic

[AskAlexSharov]

I think it must be EmptyByteSlice, not nil. Because we using Nil as indicator of “end of iteration over dbi”.

If you insert key of len 0 - it will be sorted in the beginning of dbi, not in the end. So, it behave as empty byte slice inside db, then it must be returned as empty byte slice.

What official C++ bindings return in this case?

[AskAlexSharov]

@missinglink answer from mdbx maintainer https://t.me/libmdbx/3882

[missinglink]

I think it must be EmptyByteSlice, not nil. Because we using Nil as indicator of “end of iteration over dbi”.

I'm happy to change it to []byte{}, but I'm still not clear if “end of iteration over dbi” is an issue as I believe that the key is checked for nil, not the value and, as far as I can tell, the getBytes() function is only used for values and not keys.

I'll update this PR as I'd like to get something merged which avoids the panic.

[missinglink]

updated via rebase

[missinglink]

Out of curiosity, what does [:val.mv_size:val.mv_size] do exactly? How is it different from [:val.mv_size]?

edit: agh it's called a "Full slice expression" and explicitly sets the capacity too, I learned something new today.

[AskAlexSharov]

I think it must be EmptyByteSlice, not nil. Because we using Nil as indicator of “end of iteration over dbi”.

I'm happy to change it to []byte{}, but I'm still not clear if “end of iteration over dbi” is an issue as I believe that the key is checked for nil, not the value and, as far as I can tell, the getBytes() function is only used for values and not keys.

I'll update this PR as I'd like to get something merged which avoids the panic.

when iterating over DupSort DBI - we using nil-check of value - to indicate end of sub-DBI.

[AskAlexSharov]

@missinglink - just double-confirm. You wan't merge this PR to lmdb-go, not to mdbx-go repo, yes?

[missinglink]

Correct, for lmdb not mdbx