I wasn't aware that you can get the bibtex data out of dblp without the crossrefs (although the option is clearly visible on the webpage). Thanks for the pointer! One pro for dblp: it's free!

Yes, I should have mentioned that MathSciNet requires a subscription. But my university library pays that, so I don't have to myself.

mathscinet isn't great for CS conferences that aren't theory related though (for example Data mining, KDD etc). DBLP is really the best for those. 

I gave up years ago. I just type them by hand.

ACM journals wants publisher info (why???). ACM DL more consistently (but not always) has this info... 

I didn't mention, but ACM DL also has a quirk I find irritating of using the "series" parameter for the abbreviated name of a conference. That's not what that parameter is supposed to mean.

DBLP would add/fix things if you email/tweet your suggestion to them. They are great...

Fix their whole database to change its capitalization to the right one? That seems like a lot of work.

Well, I asked for significantly more minor things...