Datastore treats str as a blob?

[jgeewax]

https://github.com/GoogleCloudPlatform/gcloud-python/blob/c3331eb80b97d5f69c243dfd421266ed22fede2c/gcloud/datastore/helpers.py#L315

This leads to us saving string values as base64 encoded blobs... What's up with that? Is that on purpose?

[jgeewax]

Looks like we went from basestr (unicode + str) to just unicode... and str is treated as a blob ad90b7b ?

[jgeewax]

Yea, looks like we did this on purpose... which seems kind of crazy that saving a regular string (ie, {'name': 'Timmy'}) ends up as a b64-encoded blob.... @tseaver ?

[dhermes]

This has been true for almost 2 whole years. See the note

Values which are "text" ('unicode' in Python2, 'str' in Python3) map to 'string_value' in the datastore; values which are "bytes" ('str' in Python2, 'bytes' in Python3) map to 'blob_value'.

Also, blobs don't get base64 encoded, they are just displayed that way in the cloud console. They get stored as byte strings, which is exactly what they are.

[jgeewax]

Hm. That's scary. Why is it that str in Python2 should convert to blob?

[dhermes]

Not scary. In Python2, str == bytes. There is no other way for us to detect that a user wants bytes vs. unicode. Storing a bytestring is much more space-efficient than storing unicode, so it's certainly something we want to enable (and not all data is strings anyhow, so we always need to enable bytestrings)

[dhermes]

$ python2.7 -c 'print(bytes)'
<type 'str'>
$ python3.4 -c 'print(bytes)'
<class 'bytes'>
$ python2.7 -c 'print(repr(b"abc"))'
'abc'
$ python3.4 -c 'print(repr(b"abc"))'
b'abc'
$ python2.7 -c 'print(repr(u"abc"))'
u'abc'
$ python3.4 -c 'print(repr(u"abc"))'
'abc'

[jgeewax]

OK, but doesn't this mean that I can't query for these results? Blobs aren't indexed are they?

It feels like if someone wants to store a blob, they should have to do something extra.... And if they store a str (which yes, happens to be == bytes), we try to store it as a string_value...

In other words, saving {'name': 'Timmy'} to datastore as a Blob value seems completely backwards. If I wanted that as a blob, I should have to do something like {'name_blob': datastore.Blob('Timmy')}...

[jgeewax]

Right -- after conversation w @dhermes it's clear that I am wrong!

Here's the deal:

• str == bytes
• If you just send off a string, we don't know the encoding, so we store it as "a string of bytes" (which happens to be called blob_value, when it might have been better named as bytes_value)
• blob_value's are indexed, which means you can query as you always did without issue
• In the Cloud Console, you'll see unencoded strings (bytes) as base-64 encoded strings. This is simply because the console itself doesn't know the encoding... They could try to guess, but that's... messy.

Long story short: use u'Timmy' if you're clear on what the encoding should be (in this case, string_value stores UTF-8 values). 'Timmy' doesn't tell Datastore what the string encoding is, and therefore Datastore just treats it as some bytes...