[Haskell-cafe] pretty-printing with fixed indentation increase per sub-structure

[Johannes Waldmann]

Dear Jeff,


> Is it possible that your size pre-calculation is inadvertently inducing quadratic behavior? 

Could be, but I don't think so, this looks pretty much linear

ghci> T.length $ L.toStrict $ renderT $ vcat $ replicate 100000 "."
199999
(0.94 secs, 468,241,960 bytes)
ghci> T.length $ L.toStrict $ renderT $ vcat $ replicate 200000 "."
399999
(1.83 secs, 935,759,344 bytes)
ghci> T.length $ L.toStrict $ renderT $ vcat $ replicate 300000 "."
599999
(2.80 secs, 1,403,276,848 bytes)

But that last expression with wl-pprint-text gives

ghci> T.length $ L.toStrict $ displayT $ renderPretty 0.5 80 $ vcat $ replicate 300000 "."
599999
(0.27 secs, 426,008,296 bytes)


Oh, and now I realize - my measurement was all wrong
since ghci interprets my code (bytecode) but has the library compiled.
(the asymptotics is still OK). When I compile my code first, I do get

ghci> T.length $ L.toStrict $ renderT $ vcat $ replicate 300000 "."
599999
(0.27 secs, 459,427,808 bytes)

and that's on par. (as it should be - I am using wl-pprint-text:SimpleDoc in the end.)


So, with that out of the way,
can we go back to discussing semantics please :-)

Have you seen 'nest' (increase indentation level)
and 'skip' (to that level, possibly with line feed) before?
Is this `hang` of the Wadler-Leijen printer, somehow?

The basic idea (p. 2)
https://homepages.inf.ed.ac.uk/wadler/papers/prettier/prettier.pdf
is the same "indentation is added only at an explicit newline,
not at the beginning of a string."
That paper was written before the "dutch layout style"
(separators at start of lines) became popular?
So it is missing the examples that I am after.
The libray https://hackage.haskell.org/package/wl-pprint-text-1.2.0.2/docs/Text-PrettyPrint-Leijen-Text.html
scares me because it mentions "three new primitives" (in extension of the paper).
But it is super fast!


Best regards, J.