david-christiansen / final-pretty-printer Goto Github PK

Using Stack with LTS-15.15 and GHC 8.8.3, I had to use allow-newer to build final-pretty-printer. Perhaps relax the upper bounds?

WARNING: Ignoring final-pretty-printer's bounds on ansi-terminal (==0.6.*); using ansi-terminal-0.10.3.
Reason: allow-newer enabled.
WARNING: Ignoring final-pretty-printer's bounds on base (>=4.3 && <4.10); using base-4.13.0.0.
Reason: allow-newer enabled.
WARNING: Ignoring final-pretty-printer's bounds on containers (==0.5.*); using containers-0.6.2.1.
Reason: allow-newer enabled.
WARNING: Ignoring final-pretty-printer's bounds on exceptions (==0.8.*); using exceptions-0.10.4.
Reason: allow-newer enabled.

We were debugging an issue with the SML version of final-pretty-printer and came across what may be a logic error in the chunk algorithm.

In short, there is the following code which determines whether to raise a fault:

    w <- measureCurLine
    n <- askNesting
    when (n + w > wmax) empty
    when (w     > rmax) empty

Observe that when nesting is enacted (through newline), spaces of the appropriate width are added to the current line (using chunk). So, in this part of chunk, it looks like this whitespace is counted twice: first in the size of the current line w, and second in the nesting n. Concretely, this results in code being too aggressively broken, when there is sufficient width for it to remain flat.

We believe that this code can be changed to the following:

    w <- measureCurLine
    n <- askNesting
    when (w > wmax) empty
    when (w - n > rmax) empty

We have tried this change out in our SML port, and it appears to give the correct result. What do you think?

--- @jonsterling and @cangiuli