An array operand can use three different indexing modes, but the boolean mask returned by a function can specify only the individual top-level items (no cells, no deeply nested stuff)
@Bubbler Yes, imo @ has unfortunate design. I wasn't quick enough to react when it was described. It should always have taken indices, even from a function right-operand, as it is trivial to convert from mask to indices (⍸), but not vice versa. The elegant @⊢ would just be @⍸ too.
CMQ: If ⍛ was added as {(⍺⍺ ⍺)⍵⍵ ⍵} wouldn't it also make sense to "overload" by allowing up to one array operand, similarly to ∘, but as opposed to ∘, it would ignore the argument on the curried side? Thus 10(+⍛1)20 would be 11 and 10(1⍛+)20 would be 21. I think this would make many trains neater.
Currently the result of f∘x or x∘f is strictly monadic, so it shouldn't interfere with existing functionality (unless some existing code depends on the erroring behavior)
@Adám Again a reason to have a monad/dyad built-in.
(A nice combo would be introducing a built-in that always errors. The sole use of J's [: is for that combo (to make the valence explicit), except for the capped fork thingy.)
@Bubbler Not hard to define, though: _MD_←{×⎕NC'⍺':⍺ ⍵⍵ ⍵ ⋄ ⍺⍺ ⍵}
@xpqz Avoid attempting to translate functions that use an operator with explicit mention of an argument in the operand (e.g. ⍣⍵). Such cannot be translated directly to tacit, though other approaches may exist.)
Ok. Out of curiosity, that dfn (Reversal (⊖) of subvectors of Y indicated by Av) -- how is it supposed to be used? It only seems to pick the items at ⍺?
@Adám Aha, I see. I read 'subvectors' to mean that the right arg was nested. In aplcart's notation, is there a way to indicate that a vector is boolean? It wasn't obvious to me that left side should be. I think that one would benefit from a TIO link.
@Adám Yes I will be explaining at one point that literature about "boolean functions" defines them as returning boolean results for boolean input, but I will stretch the definition for the sake of showing useful constructs
@RGS if you had 3 tabs open at the same time, what would happen to the rest of them when you go forward in a higher one? Which tab would be represented in the REPL?
@dzaima I think I understand your point, but the windows editor allows cycling through the various interrupted functions.
@xpqz Alright, so you are saying that if my stack was interrupted at f → g → h, I can reach f by pressing Esc on h and g and then use the debugger to step in g and h again if I need to?
Ha. In my defence -- I've not looked for it, as not needed it, but also -- even reading that now I'd never relate it to using the debugger.
Using, say, the Python debugger in VS Code, you get a nice little window showing the stack, and you can click on any stack frame and jump directly to that state. I assume that's what the Dyalog-win-IDE does, from what @RGS suggested.
@xpqz Actually, this may not work. I think it'll cut individual suspensions, not move up the stack. You may need to execute →0 for each step up. That certainly works.
More from my deepdive into namespaces and json: this, whilst achingly elegant, is also too inefficient for any practical use unless I'm being stupid (not unheard of): docs.(≢nodes.⎕NL¯9) -- docs here an array of 207 namespaces, each containing a namespace nodes of 3, 6 or 12 namespaces. The above takes me nearly 15 seconds to run.
i wonder if the runtime of these expressions just goes up unconditionally the more it's called, or if we just happened to ask them in increasing difficulty
@Adám no, hard interrupt rarely interrupts in RIDE, as it only interrupts the interpreter, thus not stopping huge formatting things as probably was the case here.
@xpqz Aha, then there's something with your specific data. If the source is json, just replacing all letters with random letters and all digits with random digits should do it.
'[aeiou]' '\w'⎕R{⍵.PatternNum:(⎕C⎕A~'AEIOU')[?21]⋄'aeiou'[?5]}'wouldn''t that be useless for anonymising though'
viehgj'y vwaj po axosobt riz usujcpiralx wyiusp
@xpqz the base JSON object has a reference count of 0 or something (i'm totally just guessing here) and Dyalog wants to GC it away so badly, but can't as a child of it is referenced somewhere
@dzaima so it's either this issue every now and then, or all namespaces can only ever be unallocated by a full GC. :/
(alternatively, don't GC on like every function call, though that's only a temporary solution and would probably just start quietly slowing things down instead of doing so loudly)
CMP: What APL glyphs should unicode have included? E.g. "Unicode should have 'left-right tack' for left and right arguments" or "Unicode should have 'intersection del diaresis' because penguins".
