Large amount of warnings lead to stack overflow when using json-messages

[michael-schwarz]

Discovered during goblint/bench#18, re-posting it here so we have a place to reference to.

A possible execution that triggers this bug ./goblint ../FFmpeg --disable ana.base.context.non-ptr --disable sem.unknown_function.spawn --set ana.thread.domain plain --enable exp.earlyglobs --set ana.base.privatization none --enable dbg.print_dead_code --set cppflags[+] -DGOBLINT_NO_BSEARCH --set cppflags[+] -DGOBLINT_NO_ASSERT --set cppflags[+] -UHAVE_INLINE_ASM --set result json-messages --set ana.ctx_insens[+] base -v

---

The stackoverflow happens inside

      Yojson.Safe.pretty_to_channel ~std:true out json

(at

analyzer/src/framework/analyses.ml

Line 278 in 36a5c5d

Yojson.Safe.pretty_to_channel ~std:true out json

).
So just using BatList.map would not cut it as the call happens inside a library.

Originally posted by @michael-schwarz in goblint/bench#18 (comment)

---

[Maybe] it's just specific to pretty_to_channel. Could be that the minimal output of to_channel doesn't suffer from the same issue.

Originally posted by @sim642 in goblint/bench#18 (comment)

[sim642]

Weird, both yojson and ppx_deriving_yojson have had issues about stack overflows, but they've also been fixed:

• stack overflow with big lists ocaml-community/yojson#47
• Stack overflow serialising long list ocaml-ppx/ppx_deriving_yojson#96

[sim642]

Oh, damn, I didn't realize that the last release of yojson (1.7.0) is three years old. The current master version wouldn't have this problem.
Still, I think to_channel should be implemented imperatively and iteratively, so it shouldn't have the issue.