@SašoŽivanović Are you sure that the examples you have from me should work?

@cfr They work when I use pgfmath-based package. Anyway, I believe I'm close: there's something about filter within sort in proof tree camau. ---Hope nobody reads ... ;-)

@SašoŽivanović It works with the version of forest I have.

root,sort by'={{O}{y},{Ow1+d}{x}{-##1}},sort'={filter'={descendants}{{OO!&}{proof tree rhifo}{proof tree phantom}}}

@cfr Something is mightily weird here. So there is this pgfmath if: if={y()<y("!back")}. In my version, the false clause is never reached! Can you check with yours? (Just put typeout=TRUE and typeout=FALSE in there.)

@SašoŽivanović That is no longer in my version. Do you want me to pull the version I sent you? If so, can you let me know the revision? I think it is 4742, but I'm not certain.

Please do. I've got 4688.

Sorry, my bad: it does get there. No need to pull.

I changed something to test and forgot to unchange it.

@SašoŽivanović OK. So this is not the last version I sent you, right?

@cfr No, I'm working on the one that is processified ... oh, that one is, too. And it's still in the email. And I promised to take a look. Oh.

Actually, I asked if I can take a look. Oh my.

@SašoŽivanović You might as well have the current version, in that case, so I've just sent you an update.

@SašoŽivanović It's not a big deal. I was only curious why something didn't work, I think.

@cfr Thanks. And all nicely fixed to the changed process interface!

@SašoŽivanović I thought it would be easier. It doesn't have the y()... though.

@SašoŽivanović Well, strictly speaking you asked for prooftrees.sty + MWE, but only because I was bugging you.

@cfr Yes and I truly wanted to look at it (and I will!) but I had to call it a day ... I'm a night person, but catcodes won! ;-) And today, I beat them!

@cfr I get the same original error. Cannot make a sibling step somewhere. That's how the y story started ...

@SašoŽivanović No problem, really.

@cfr And this version you sent me, it works for you?

@SašoŽivanović I'm just double-checking.

@cfr Me too. Works on my master branch.

@SašoŽivanović Yes. The prooftrees-egs.tex compiles with the prooftrees.sty I just sent you and the forest.sty you sent me last.

And the output looks OK as far as I can tell.

Since it worked yesterday and I've not touched any of it since, this is good!

@cfr Yes the output of that one was the same as v2.0.2 --- even diffpdf reported them identical.

@cfr ;-)

@SašoŽivanović The same may not be true for intermediate versions as I introduced some bugs to do with alignment. But I think they are fixed now.

@cfr There is something, though, that the new version doesn't have: side by side pgfmath and .process ... to switch between the two.

It seems that I'll have to find this bug with the old one.

@SašoŽivanović Yes, sorry. I had both in initially but tended to clean up the comments when I got things working as it made it easier to see what I was doing.

@cfr That's ok. For my purpose, It doesn't really matter which one I'm working with.

@SašoŽivanović I can send you a diff if you want.

No need. Actually, I just figured out the y business. It's bogus, anyway, it happened pgfmath expands stuff. So I was misled ;-(

@SašoŽivanović It was a stupid suggestion, anyway. You have both versions so you could diff it yourself.

@cfr ;-)

@SašoŽivanović Note that the earlier version won't work with the non-minimal example from my email because it doesn't support splitting the tree.

@cfr I'll check that example with the new one. And i'm getting closer to understanding this here. It might not even be a bug!

@SašoŽivanović More of a gardener's friend ?

@cfr Sort of. I mistranslated something from your code to .process. I thought the two logics were equivalent, but they are not.

@SašoŽivanović Not always. At least, that's how I found it.

@cfr GOT THE BUG!!!

Not all arrays start with 0 ...

@SašoŽivanović ??

@SašoŽivanović ;)

@cfr You know, the gardener's friend!

@SašoŽivanović I meant the arrays comment but I replied to the wrong one.

@cfr While my shiny new arrays do initially start at 0, they can later in life start anywhere ... don't mind.

@cfr Bottomline is: your examples compile. I tested performance. It is the same as with the old .process. Both are meager 10% better than pgfmath.

@SašoŽivanović Ah. Well, proof trees come to those who wait....

@cfr :-) ;-(

@SašoŽivanović Well, now you can just blame my design. Or TeX. ;)

Though performance is actually quite consistent with my dry tests ... if I had 20-30% there, small wonder I'm down to 10 in a real application.

@cfr I go for our beloved TeX! :D

@SašoŽivanović I wouldn't know what to expect one way or the other. This is well beyond me.

@cfr But I managed to get additional 3% (measured on my docs) by optimizing option/register access code!

Whatever I do, forest's gonna be the slowest package in the universe!

@SašoŽivanović No prooftrees will be slower.

@cfr Yes but you're cheating! ;-))))

@SašoŽivanović You obviously haven't used TikZ's text decorations as much as you should.

@cfr Oh.

@SašoŽivanović Specifically, text effects along path. text along path is not bad. But text effects along path typesets every character in its own node ....

Even the manual says 'due to the number of computations involved, this is quite a slow decoration'.

@SašoŽivanović The process stuff is also easier to use, I think, than the pgfmath wrapping. Especially if you want something expanded or whatever. And it is surely more flexible. Also, 10% is hardly insignificant. For documents with lots of trees, especially. And TikZ itself is slow. Even before you use it, just loading it is significant.