Fix wrong Content-Length with non ASCII-8BIT strings.

[meh]

I was implementing a proxy and noticed I was getting a truncated result, after some time thinking it was my fault I thought about the Content-Length being wrong, and apparently evma_httpserver uses #length instead of #bytesize.

Putting #bytesize everything works properly.

Sorry for the change in the gemspec but it wasn't building with this error:

ERROR: While executing gem ... (Gem::InvalidSpecificationException)
cert_chain must not be nil

I also faced another problem with header names, it would be really best if you used the common names and not weirdly different names, for instance: Content-Length instead of Content-length. If it's wanted I could write a patch for this and send another pull request. (EDIT: #10 this does what I meant)

Thanks.

[tmm1]

Shouldn't this be content instead of @content, to prevent calling nil.bytesize

[meh]

Nope, @content presence is already checked above, there's no chance for it to be nil.

[rcarvalho]

Thanks for that fix, guys. Is anyone still taking ownership of this code? I see that they still haven't merged your fix in. You wouldn't have happened to publish a new version to rubygems.org would you? I guess I could just fork and merge in your changes myself.

[jay2u]

This is actually still an issue. We are using a monkey patch to fix this at the moment (after spending quite some time finding the origin of this problem) only to find out there was already a pull request for this...

Any reason this cannot be merged into the master branch?

Don't get me wrong. Thanks for a great piece of software otherwise. Brilliant stuff...

[halida]

The newest gem version 0.2.1 still has this problem, consider release new version?

[philayres]

@halida - I just found this too. I worked around the issue directly in my code with:

response.content = content.force_encoding(Encoding::ASCII_8BIT)

Its a shame that the current version of the gem doesn't handle this. Time for a new version.

[jay2u]

@philayres Yes that would work but only with ASCII_8BIT. The issue here, amongst other things, is trying to avoid unicode body payloads becoming truncated.