Created attachment 189193 [details] [review]
proposed (but untested) patch for adding use_agent flag

This patch against gmime 2.4 is untested as i appear to be unable to build from source at the moment.

I'm also not sure if it introduces an incompatible ABI change due to the modification of the _GMimeGpgContext struct (does the leading underscore mean it it's a private definition despite being in a public header file?) -- if it's an ABI bump, that's clearly a no-go for the stable branch.

i'd be happy to see this on the 2.5 branch as well, of course.

I note that one oddity here is that FALSE simply means "don't supply an additional --use-agent flag" -- so if the user still has use-agent set in ~/.gnupg/gpg.conf, they will still talk to the agent.

The proposed patch has no way to indicate "supply --no-use-agent as an argument", which would actually allow the gmime invocation to explicitly prohibit gpg from talking to any running agent.

I actually think this asymmetry is fine, since i don't see a good argument for avoiding talking to an agent if one is present (and the upcoming gpg 2.1 actually requires the use of an agent, so --no-use-agent would be crippling for any crypto invocation that needed access to secret credentials).

I agree about the --no-use-agent thing.

As far as ABI breakage, yes, unfortunately it breaks ABI :(

struct _GMimeGpgContext gets typedef'd to GMimeGpgContext, so it is public.

I've modified the patch a bit. It was only setting use_agent for sign+encrypt but not sign nor decrypt, so I made it pass that option along for those modes as well.

I also ported your patch to 2.5.x and committed it there.