On 04/06/2010, at 11:59 PM, Manuel Pégourié-Gonnard wrote: > This explains the difference in the output format. Unfortunately, I'm not sure > how to reliably handle this.
Easy :) Update my version. Unfortunately MacPorts is a big pain to use and everything must be compiled from scratch. And ImageMagick uses *a lot* of dependencies. Hence the delay on this message. I've now updated the Makefile for the newer version of ImageMagick but the numbers are all a hack. Maybe now it will work? BTW, the reason that you get "rounding errors" or whatever I was calling them, is that the PDFs must first be converted to PNG (with a specific resolution) before being compared. Imperceptible changes in glyph placement can affect how the image is discretised, and you can end up with a line of pixels, say, (or sometimes just a single pixel) that have ended up being generated differently. There might be some antialiasing effects that're compounding the problem but I haven't looked into ImageMagick enough to know if that's the case (and if so, how to turn it off). >>> Or maybe we can combine both modes: ship the .safe.pdf files (I think they >>> are >>> valuable) and allow a (local) regression-testing mode. It should be doable >>> by >>> adjusting the location of the reference files depending on the target. >> >> That's a good idea. >> > But it makes more work for you :-) Remind me if I forget, but that doesn't seem like toooooo much work. -- Will
