Over the last two days, I actually went through every page of the local help document, and really tried to learn how everything works. I understand DxO quite a bit better than I had previously. I still need to get my mind wrapped around the best integration of DxO in my own workflow, but I feel that I now understand all of the settings, and how the designers anticipated they would be used by customers.
Prior to studying the manual, I had been operating DxO in a less than optimal manner. Two examples:
First, you have to actually process an image to get a feel for what the final corrections will look like. Obvious, I know, but I had a habit before of making tweaks at 100%, then looking at the "fit to window" image without realizing DxO doesn't even simulate their final corrections at anything under 75% magnification. That can make a big difference when evaluating contrast settings.
Second, "DxO Lens Sotness" and "Unsharp Mask" shouldn't be used together. Unsharp Mask is really intended just for images that were shot with a lens that has no DxO module, such as my Sigma macro. Using both corrections produces ugly artifacts, which I had previously attributed to DxO's logic. Oops. From now on, I'll look at DxO Lens Sharpness as an analog to Bruce Fraser "capture sharpening." If in image needs more than that, I'll be doing it in Photoshop.
Incidentally, I only found one example of a feature that was present in the documentation, but not in the functioning software (a "stack" button in the project window, which isn't really needed anyway, since there's a stack option on the right-click menu). Previous versions were sometimes out of sync with their documentation, but this one doesn't seem so bad.
The one remaining serious annoyance (for me) is that DxO refuses to read the focus distance information embedded in the EXIF for 40D images. Not a problem for most landscaps shots, which are focused at infinity anyway. But the corrections do look different at close focus, and it sure would be a lot better if DxO would just pick up the data that's already in the image. DPP can, of course.
My next step is to run an experiment to compare raw shots converted with DxO, Lightroom 1.4, DPP and ACR. I'll probably do that in phases. First will be an evaluation of noise/sharpness, then geometry/distortion, and finally light and color. To some extent, those qualities end up overlapping, of course. But I feel I need to break my comparison into controllable pieces, or I'll never have anything other than a vague feel for which is "better" overall.
Looking at the images I used while studying the manual, I can say that 5.2 produces a really nice balance between noise and sharpness, at least for the few old 20D shots I was working with. The default "DxO Lighting" corrections have been pretty good so far, and if anything are usually a little conservative. The color balancing tools are a hair less intuitive than their Lightroom 1.4 equivalents, but they seem to be pretty effective.
I also like the way you can set up concurrent versions of the same image for the same project, and then run mutliple output files for each in a single process. Let's say you have an image that might look good in both color and black and white. You can just make a duplicate selection of the same raw file in the project window, and set them up with entirely different color settings. Then when you go to the "process" tab, you can define multiple types of outputs. Maybe a 16-bit TIFF for further editing in Photoshop, a small jpg for a web site, and a larger jpg for an e-mail attachment. Until you've actually processed those final outputs, you've never duplicated any files. You can do similar things with Lightroom, but you can't just process a whole set of different kinds of output files with one step (or at least I can't.)