Stack overflow in js_of_ocaml #26

aantron · 2017-07-06T01:03:52Z

Hi. I'm not sure what the state of this discussion is, but I can report on my recent experience of using markup with js_of_ocaml (wrt bucklescript, I would very much like to use jsoo and not BS, and do not really care for the size of the produced js file).

Essentially it is very easy to run into a stack overflow, when using markup compiled to js. I suspect this is due to the CPS used in the implementation, which I assume is hard for jsoo to optimize into tail-recursion.

One solution I see would be to manually trampoline in markup, which is a somewhat invasive > change...

Repro case: https://paste.isomorphis.me/O8j

and confirmed in a subsequent comment.

Armael · 2017-07-06T01:05:45Z

See http://ocsigen.org/js_of_ocaml/2.8/manual/tailcall .

Armael · 2017-07-06T01:06:30Z

As far as I know, CPS is not optimized into tailcalls by js_of_ocaml.

Armael · 2017-07-06T02:39:45Z

@hhugo Is there something that could be done on jsoo's side? Or is the only solution to implement trampolining in markup itself?

Jo-Blade · 2024-07-25T07:45:25Z

Hello, is there anything new about this problem today ? Do you know any alternative for xml or http parsing that would work with jsoo ? I can confirm the issue still persist with markup 1.0.3 and js_of_ocaml 5.8.1

In my case, the stack overflow occurs even with html string as simple as:

let example_html = {|
<!DOCTYPE html>
<html lang="en">
  <body>
    <script type="text/javascript" src="example.js"></script>
  </body>
</html>
|}

let _ = Soup.parse example_html

aantron · 2024-07-25T08:06:39Z

Hi, no progress as of now. Could you give your complete workflow, if it's not too much work to simplify, so I can readily observe the problem? Some sample program, its Dune files, commands you are running? Is this in Node or a browser?

For parsing XML you could try Xmlm. For HTML there is a very non-standards-compliant HTML parser in ocamlnet, but I doubt ocamlnet itself is compatible with js_of_ocaml, though I recall previously extracting the parser from that library. You might also be able to get by by treating well-enough-formed HTML as if it were XML and using an XML parser.

But otherwise, it might be best to adapt Markup.ml to be useful in js_of_ocaml.

hhugo · 2024-07-25T08:34:56Z

@aantron, just in case it's useful, here is how I adapted jsonm for jsoo : dbuenzli/jsonm#20

hhugo · 2024-07-25T08:44:14Z

dune

(executable
  (name main)
  (modes js)
  (libraries lambdasoup)
  (js_of_ocaml (flags --pretty))
)

main.ml

let example_html =
  {|
<!DOCTYPE html>
<html lang="en">
  <body>
    <script type="text/javascript" src="example.js"></script>
  </body>
</html>
|}

let _  = Soup.parse example_html

commands:

dune build main.bc.js
node _build/default/main.bc.js

     throw err;
     ^

RangeError: Maximum call stack size exceeded
    at caml_call1 (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:39429:15)
    at iterate (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:41333:26)
    at caml_call1 (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:39429:15)
    at /home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:40406:26
    at caml_call3 (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:26155:15)
    at f (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:26184:13)
    at caml_call3 (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:26155:15)
    at f (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:26184:13)
    at caml_call3 (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:26155:15)
    at next (/home/hugo/js_of_ocaml/_build/default/mark/main.bc.js:26194:12

Adding Error.stackTraceLimit = Infinity; at the top of the js file can give you the whole trace.

Note that using the release profile works in that specific example, most probably because jsoo is able to inline more (e.g. cross module inlining)

dune build --profile release main.bc.js
node _build/default/main.bc.js

Jo-Blade · 2024-07-25T14:53:42Z

Hi, no progress as of now. Could you give your complete workflow, if it's not too much work to simplify, so I can readily observe the problem? Some sample program, its Dune files, commands you are running? Is this in Node or a browser?

For parsing XML you could try Xmlm. For HTML there is a very non-standards-compliant HTML parser in ocamlnet, but I doubt ocamlnet itself is compatible with js_of_ocaml, though I recall previously extracting the parser from that library. You might also be able to get by by treating well-enough-formed HTML as if it were XML and using an XML parser.

But otherwise, it might be best to adapt Markup.ml to be useful in js_of_ocaml.

In fact the code is a http parser that fetch the code of a website and extract all links to display them in a simple cli. So basically I use lambdasoup to look for all "" balise, extract their "href" attribute and the rendered text (because it can contains a lot of useless div and I'm only insterested in the rendered visible text). Following is the failing snippet :

(* s is the string response from cohttp-lwt-jsoo *)
fun s ->
Soup.parse s $? "div[role=main]" |>> fun div ->
(* Get all links of the current div *)
div $$ "a" |> Soup.to_list |> fun l ->
`Ok
(* Convert all links and keep only non "None" results *)
(List.fold_left
(fun acc x ->
  match
    do_conversion (Soup.attribute "href" x) (Soup.trimmed_texts x) with
        | None -> acc
        | Some y -> y :: acc)
      [] l))

As I'm fetching website content, the target platform is nodejs (because in web browser it would trigger cors policy).

To be honest, it's not a big deal if I cannot use js_of_ocaml as it's was not even originally planned for this project. It's just a little disappointing because I was expecting js_of_ocaml to be an easy way to distribute my program for windows users (for example) without extra work. I've already done some tests so it's really this snippet that cause the all program to fail, if I replace it with a fake returned value, the program compiles and works.

v-gb · 2024-12-18T21:42:57Z

Not sure if there's interest for upstreaming this functionality, but I made markup work in javascript in this branch: master...v-gb:markup.ml:js. It uses trampolines, which is a fairly lightweight change.

I had tried compiling with the jsoo flag that enables cps transformation, but (from memory) it made the code much slower to compile, much slower to run, much bigger, and didn't execute correctly when enabling optimizations. So I much prefer the trampoline version.

It's not very fast, but I think that's due to markup's one-character-at-a-time approach, rather than the trampoline overhead.

aantron mentioned this issue Jul 6, 2017

Compile to JS and publish in NPM #22

Open

Armael mentioned this issue Jul 17, 2017

Stack overflow when processing big html elements Armael/tyxml-translator#8

Open

aantron added this to the 0.8.2 milestone Aug 17, 2019

aantron mentioned this issue Sep 20, 2020

usage from js_of_ocaml aantron/lambdasoup#36

Closed

aantron removed this from the 0.8.2 milestone Oct 22, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stack overflow in js_of_ocaml #26

Stack overflow in js_of_ocaml #26

aantron commented Jul 6, 2017

Armael commented Jul 6, 2017

Armael commented Jul 6, 2017

Armael commented Jul 6, 2017

Jo-Blade commented Jul 25, 2024 •

edited

Loading

aantron commented Jul 25, 2024

hhugo commented Jul 25, 2024

hhugo commented Jul 25, 2024

Jo-Blade commented Jul 25, 2024

v-gb commented Dec 18, 2024 •

edited

Loading

Stack overflow in js_of_ocaml #26

Stack overflow in js_of_ocaml #26

Comments

aantron commented Jul 6, 2017

Armael commented Jul 6, 2017

Armael commented Jul 6, 2017

Armael commented Jul 6, 2017

Jo-Blade commented Jul 25, 2024 • edited Loading

aantron commented Jul 25, 2024

hhugo commented Jul 25, 2024

hhugo commented Jul 25, 2024

Jo-Blade commented Jul 25, 2024

v-gb commented Dec 18, 2024 • edited Loading

Jo-Blade commented Jul 25, 2024 •

edited

Loading

v-gb commented Dec 18, 2024 •

edited

Loading