search

ghcjs / jsaddle

JavaScript interface that works with GHCJS or GHC
116 stars 62 forks source link

Can't build `jsaddle-webkit2gtk` #85

Open vaibhavsagar opened 5 years ago

vaibhavsagar commented 5 years ago

It looks like GI.JavaScriptCore and GI.WebKit2 have changed in incompatible ways:

Building library for jsaddle-webkit2gtk-0.9.6.0..
[1 of 1] Compiling Language.Javascript.JSaddle.WebKitGTK ( src/Language/Javascript/JSaddle/WebKitGTK.hs, dist/build/Language/Javascript/JSaddle/WebKitGTK.o )

src/Language/Javascript/JSaddle/WebKitGTK.hs:68:38-54: error:
    Module ‘GI.JavaScriptCore’ does not export ‘GlobalContext(..)’
   |
68 | import GI.JavaScriptCore (Value(..), GlobalContext(..))
   |                                      ^^^^^^^^^^^^^^^^^

src/Language/Javascript/JSaddle/WebKitGTK.hs:77:9-32: error:
    Module ‘GI.WebKit2’ does not export ‘javascriptResultGetValue’
   |
77 |         javascriptResultGetValue, javascriptResultGetGlobalContext,
   |         ^^^^^^^^^^^^^^^^^^^^^^^^

src/Language/Javascript/JSaddle/WebKitGTK.hs:77:35-66: error:
    Module
    ‘GI.WebKit2’
    does not export
    ‘javascriptResultGetGlobalContext’
   |
77 |         javascriptResultGetValue, javascriptResultGetGlobalContext,
   |

To reproduce, use this default.nix:

let
  nixpkgs-src = builtins.fetchTarball {
    url = "https://github.com/NixOS/nixpkgs/archive/6a0d2ff7c1d024914a3570b85f1c88df8930b471.tar.gz";
    sha256 = "06q10786lj7yig6spzsz1zflcjqxjj11d5qsr305jln34wndlj27";
  };
  jsaddle-src = builtins.fetchTarball {
    url = "https://github.com/ghcjs/jsaddle/archive/98a00b334b0ce62bf6bd3a4af682b25a8ea28193.tar.gz";
    sha256 = "057a4imgyq4aabsqhlv9zd3z855cwii2qq8i5hmfnfmc9wcx403l";
  };
  pkgs = import nixpkgs-src {};
  haskellPackages = pkgs.haskellPackages.override {
    overrides = self: super: {
      jsaddle = self.callCabal2nix "jsaddle" (jsaddle-src + "/jsaddle") {};
      jsaddle-webkit2gtk = self.callCabal2nix "jsaddle-webkit2gtk" (jsaddle-src + "/jsaddle-webkit2gtk") {};
      haskell-gi-overloading = self.callHackage "haskell-gi-overloading" "0.0" {};
    };
  };
in haskellPackages.jsaddle-webkit2gtk
guibou commented 5 years ago

Current nixpkgs uses webkit-gtk-2.22. I wonder if jsaddle-webkit2gtk is not based on previous version (and API?)

develop7 commented 5 years ago

This affects me too. Seems to be caused by the lack of said GlobalContext in gi-javascriptcore-4.0.16 (it is present in 4.0.15 though!)

chessai commented 5 years ago

I am also experiencing this.

amanbhurji commented 5 years ago

Me too. Using 4.0.15 only changes the error to -

GI/JavaScriptCore/Objects/Exception.hs:120:1: error:
    Could not find module ‘GI.GObject.Objects.Object’
ryantrinkle commented 5 years ago

This builds (and so do lots of transitive dependent packages), but I haven't tested it: https://github.com/ryantrinkle/jsaddle/tree/no-global-context

develop7 commented 5 years ago

Thank your for looking into it, @ryantrinkle. I've built a "hello world" reflex app (built with Stack though, not Nix) using jsaddle from above branch; it shows empty window then segfaults in libjavascriptcoregtk:

$ coredumpctl gdb
           PID: 27757 (reflex-stack-ex)
           UID: 1000 (develop7)
           GID: 100 (users)
        Signal: 11 (SEGV)
     Timestamp: Mon 2019-04-29 22:52:36 +03 (19min ago)
  Command Line: /home/develop7/projects/reflex-stack/.stack-work/install/x86_64-linux-tinfo6/lts-12.26/8.4.4/bin/reflex-stack-exe
    Executable: /home/develop7/projects/reflex-stack/.stack-work/install/x86_64-linux-tinfo6/lts-12.26/8.4.4/bin/reflex-stack-exe
 Control Group: /user.slice/user-1000.slice/session-2.scope
          Unit: session-2.scope
         Slice: user-1000.slice
       Session: 2
     Owner UID: 1000 (develop7)
       Boot ID: 34ad9b77fbf44d83a582ae08838287e4
    Machine ID: f7d12f47fb5c49979f8d99a60666bab2
      Hostname: t-rex
       Storage: /var/lib/systemd/coredump/core.reflex-stack-ex.1000.34ad9b77fbf44d83a582ae08838287e4.27757.1556567556000000.lz4
       Message: Process 27757 (reflex-stack-ex) of user 1000 dumped core.

                Stack trace of thread 27757:
                #0  0x00007fae3b35b763 _ZN3JSC12JSLockHolderC2ERNS_2VME (libjavascriptcoregtk-4.0.so.18)
                #1  0x00007fae3a9e7bad JSValueToStringCopy (libjavascriptcoregtk-4.0.so.18)
                #2  0x00000000004a9de6 n/a (/home/develop7/projects/reflex-stack/.stack-work/install/x86_64-linux-tinfo6/lts-12.26/8.4.4/bin/reflex-stack-exe)
…skip…
Program terminated with signal SIGSEGV, Segmentation fault.
#0  std::__atomic_base<unsigned int>::operator++ () at /usr/include/c++/8/bits/atomic_base.h:296
296       { return __atomic_add_fetch(&_M_i, 1, memory_order_seq_cst); }
[Current thread is 1 (Thread 0x7fae3a3bdb00 (LWP 27757))]
(gdb) bt
#0  0x00007fae3b35b763 in std::__atomic_base<unsigned int>::operator++() () at /usr/include/c++/8/bits/atomic_base.h:296
#1  0x00007fae3b35b763 in WTF::ThreadSafeRefCountedBase::ref() const () at DerivedSources/ForwardingHeaders/wtf/ThreadSafeRefCounted.h:43
#2  0x00007fae3b35b763 in WTF::refIfNotNull<JSC::VM>(JSC::VM*) () at DerivedSources/ForwardingHeaders/wtf/RefPtr.h:38
#3  0x00007fae3b35b763 in WTF::RefPtr<JSC::VM, WTF::DumbPtrTraits<JSC::VM> >::RefPtr(JSC::VM*) () at DerivedSources/ForwardingHeaders/wtf/RefPtr.h:57
#4  0x00007fae3b35b763 in JSC::JSLockHolder::JSLockHolder(JSC::VM&) () at ../Source/JavaScriptCore/runtime/JSLock.cpp:64
#5  0x00007fae3a9e7bad in JSValueToStringCopy() () at ../Source/JavaScriptCore/API/JSValueRef.cpp:432
#6  0x00000000004a9de6 in  ()
#7  0x0000000004b17d68 in  ()
#8  0x0000000001e93ed8 in  ()
#9  0x0000000000000000 in  ()

Hope that helps making further progress.

ryantrinkle commented 5 years ago

@develop7 Ah, looks like I guessed wrong about GlobalContext and Context being sufficiently interchangeable. Maybe @hamishmack has an idea?

hamishmack commented 5 years ago

I think the first step is to create a Haskell library to wrap the new jsc-glib library that is now in webkitgtk. This new haskell library would take the place of the webkit2-javascriptcore that jsaddle-webkit2gtk currently uses.

Unfortunately I don't think jsc-glib has gobject introspection (though I would love to be proved wrong on that). Hopefully once we have that we can replace the code that currently calls C functions like JSValueToStringCopy with code that calls the new C functions like jsc-value-to-string-as-bytes.

I think we should avoid any dependency on gtk2hs-buildtools and instead use a more standard tool like hsc2hs. The gi-cairo-render package might be a useful place to look for ideas and @cohomology might have some good tips on how to create a haskell bindings for a C library with good haskell-gi interop.

Another option is to add FFI imports for just the jsc-glib functions we need in line in the jsaddle-webkit2gtk code (that might be less work, but it would suck a little).

ali-abrar commented 5 years ago

I think #96 fixes this by avoiding global context entirely. I was able to build and run reflex-todomvc with that patch.

vaibhavsagar commented 5 years ago

This is so exciting! Maybe we can finally get servant-client-jsaddle over the finish line :smile:.

develop7 commented 5 years ago

I think #96 fixes this by avoiding global context entirely. I was able to build and run reflex-todomvc with that patch.

yup, that does it — just got reflex "hello world" running in webkitgtk. Thanks, man!