UTF-8 conversion on ruby 1.9.1

[jpzwarte]

Hi,

When using your excellent sqlserver adapter with ruby 1.9.1, uploading a binary (PNG in my case) will fail in the binary_to_string method with the following error:

ArgumentError (invalid byte sequence in UTF-8):
/Users/jeroen/.bundle/ruby/1.9.1/gems/activerecord-sqlserver-adapter-2.3.4/lib/active_record/connection_adapters/sqlserver_adapter.rb:61:in binary_to_string' /Users/jeroen/.bundle/ruby/1.9.1/gems/activerecord-sqlserver-adapter-2.3.4/lib/active_record/connection_adapters/sqlserver_adapter.rb:70:intype_cast'

I looked at the source of activerecord and found that the sqlite adapter has an apparent fix/workaround for this:

def binary_to_string(value)
  value = value.dup.force_encoding(Encoding::BINARY) if value.respond_to?(:force_encoding)
  # etc.

Adding the above line to the sqlserver adapter fixes the problem for me.

The sqlite adapter also uses this for string_to_binary:

def string_to_binary(value)
  value = value.dup.force_encoding(Encoding::BINARY) if value.respond_to?(:force_encoding)
  # etc.

Could you incorporate these changes in your adapter and release a new version?

Thanks,
Jeroen

[metaskills]

Is the dup on this necessary? If that was a big file would it not dup that big object and bloat memory? Just curious.

[metaskills]

Also, are you saying this line is needed for both methods?

[jpzwarte]

Not sure, i basically just looked for a workaround. I guess though that dup'ing the value keeps the original intact. Otherwise the value would change in the model?

[jpzwarte]

Yes, iirc it is needed for both methods.

[metaskills]

I'm waiting for 1.9.2 before settup up my ruby switcher again. So I'll just take your word that since no tests are failing under 1.8 that this is OK. I'll push the commit up soon, a release will be in a few days or so.

[jpzwarte]

rvm is a breeze to get 1.9.2 running, i recommend using that approach.