Fix stack overflow on large string constants

[gpetiot]

Fix #1561, no diff with test_branch.

[emillon]

how long does this test take to run? also I'm wondering if it makes sense to generate the string instead of storing it?

[gpetiot]

It's almost instant for me, and I'm running on a slow vm

[emillon]

What do you think of generating this test, either as a dune rule or as a cram test?

[gpetiot]

It looks to be difficult to generate a string that is already formatted (-n 1), maybe that would be enough to have a unit test for list_pn instead of testing the absence of exception of the ml file?

[emillon]

ha yes I didn't realize the file was already format. let's scratch that idea and keep the huge file. the unit test is complementary for that, it won't check the resources, just the result. so it all seems good!

[gpetiot]

Yes I can reproduce it with echo "let _ = \"$(printf '%*s' 300000 | sed 's/ /_ _/g')\"" | ocamlformat - --impl -o /dev/null --debug but I don't get any trace

my bad that wasn't the bytecode one, the backtrace is made of countless lines of

Called from Ocamlformat_lib__Fmt.T.($).(fun) in file "lib/Fmt.ml", line 50, characters 27-30

ending with another stack overflow, on ( $ ) this time @emillon

[emillon]

OK I'll try if this can be fixed with a different encoding.

[emillon]

I've got a fix for this, I'll open a PR

[emillon]

Done in #1568. Then you can write:

       let l =
         let rec aux (prev, acc) = function
          | [] -> acc
           | [xI] -> aux (xI, (Some prev, xI, None) :: acc) []
           | xI :: (xJ :: _ as xJN) ->
               aux (xI, (Some prev, xI, Some xJ) :: acc) xJN
         in
         aux (x1, [(None, x1, Some x2)]) x2N
       in
      List.rev_map l ~f:(fun (prev, x, next) ->
          lazy_ (fun () -> pp ~prev x ~next) )
      |> sequence

[gpetiot]

Thanks, I rebased this one, the CI is now ok!

[emillon]

I added a test in #1566.

[emillon]

This is still segfaulting on ppc64 though, interesting