[geek1011]

Originally Posted by davidfor

You mean you don't want the dictionary problems that you extensively described fixed?

The only purpose of this release is to fix the issues with "prefix_exceptions" in the dictionary.

I didn't look at the dictionary stuff specifically (those kinds of changes which only affect individual instructions in existing functions are the hardest to notice, and I would have overlooked this too since it's almost be too good to be true to have that fixed so quickly). But yes, I did want those fixed if it was possible and if it was actually an issue rather than not using it correctly (after all, I did figure everything out based on the assembly and some educated guesses for relevant test cases, since there aren't yet any available official v3 dictionaries using prefix_exceptions). In that case, I'm quite pleased with Kobo.

Thanks for the pointer about what changed. It'll save me a lot of time, both for figuring out what changed, and for when I test dictionaries expecting it to do the same thing. Remember that I'm looking at a disassembly, not the original source code, and that doesn't cleanly diff or do things in the most logical manner.



I think this will become my favourite release so far, as almost everything I've personally wanted fixed in the firmware has now been fixed, and I haven't discovered any major issues/regressions with the features I use so far.

Edit: I'll look through the code again and post a more details summary of the changes when I have time (either tomorrow night or two days after that).

Edit 1: I just took a quick peek at the most likely places, and I think I've found the changes. I'll come up with test cases later.

Edit 2: For anyone interested, the most important part is at 0x5a61ac in 15672 and 0x5a643c in 15676. I'll also need to debug one of my scripts, since it should have noticed a side effect of this change automatically.

Edit 3: The bug was in my parsing of the JSON diff of the output of one of my tools (it didn't detect single-character changes properly).

Edit 4: I can confirm this fixes the default behavior for my main test cases . That's all I'm going to check for today. Tomorrow, I'll look at exactly how it does it and what changed.