Feature #5007: Proc#call_under: Unifying instance_eval and instance_exec

• Status: Assigned
• Assignee: matz (Yukihiro Matsumoto)
• Target version: -

[judofyr (Magnus Holm)]

I'm proposing a method called Proc#call_under (the name could be
discussed) which both unifies instance_eval and instance_exec, and makes
it possible to call a Proc with a block and a scope:

Proc#call_under(self, *args, &blk):

proc { self }.call_under(1) # => 1

proc { |a| self + a }.call_under(1, 2) # => 3

proc { |&b| self + b.call }.call_under(2) { 2 } # => 4

[mame (Yusuke Endoh)]

• Assignee set to matz (Yukihiro Matsumoto)
• Status changed from Open to Assigned

[trans (Thomas Sawyer)]

I don't see how this solves the case presented in #6298.

Given:

  class BlockCollection
    def initialize(*procs)
      @procs = procs
    end
    def to_proc
      procs = @procs
      Proc.new{ |*a| procs.each{ |p| p.call_under(self, *a) } }
    end
  end

What is self here, or what should it be? Such that:

  module M
    A = Proc.new{ self }
  end

  bc = BlockCollection.new(A)

  bc.call  #=> M

  'F'.instance_eval(&bc)  #=> 'F'

[mame (Yusuke Endoh)]

• File proc_call_under.patch proc_call_under.patch added

Hello,

I made a proof-of-concept patch.
Not tested yet. Please try it and find a bug.
It (and some related functions) seem to need some refactoring work
because it calls directly invoke_block_from_c which is very internal
function.

  $ ./miniruby -e '
    p proc { self }.call_under(1)
    p proc { |a| self + a }.call_under(1, 2)
    p proc { |&b| self + b.call }.call_under(2) { 2 }
  '
  1
  3
  4

diff --git a/proc.c b/proc.c
index d44e8d8..7ad490e 100644
--- a/proc.c
+++ b/proc.c
@@ -567,6 +567,22 @@ proc_call(int argc, VALUE *argv, VALUE procval)
     return vret;
 }
 
+VALUE rb_proc_call_under(VALUE procval, VALUE under, VALUE self, VALUE values);
+
+static VALUE
+proc_call_under(int argc, VALUE *argv, VALUE procval)
+{
+    VALUE self, klass, values;
+    rb_scan_args(argc, argv, "1*", &self, &values);
+    if (SPECIAL_CONST_P(self)) {
+	klass = Qnil;
+    }
+    else {
+	klass = rb_singleton_class(self);
+    }
+    return rb_proc_call_under(procval, klass, self, values);
+}
+
 #if SIZEOF_LONG > SIZEOF_INT
 static inline int
 check_argc(long argc)
@@ -2183,6 +2199,7 @@ Init_Proc(void)
     rb_define_method(rb_cProc, "[]", proc_call, -1);
     rb_define_method(rb_cProc, "===", proc_call, -1);
     rb_define_method(rb_cProc, "yield", proc_call, -1);
+    rb_define_method(rb_cProc, "call_under", proc_call_under, -1);
 #endif
     rb_define_method(rb_cProc, "to_proc", proc_to_proc, 0);
     rb_define_method(rb_cProc, "arity", proc_arity, 0);
diff --git a/vm_eval.c b/vm_eval.c
index 6c26b97..562a215 100644
--- a/vm_eval.c
+++ b/vm_eval.c
@@ -1262,6 +1262,30 @@ yield_under(VALUE under, VALUE self, VALUE values)
     }
 }
 
+static inline VALUE
+invoke_block_from_c(rb_thread_t *th, const rb_block_t *block,
+		    VALUE self, int argc, const VALUE *argv,
+		    const rb_block_t *blockptr, const NODE *cref);
+
+VALUE
+rb_proc_call_under(VALUE procval, VALUE under, VALUE self, VALUE values)
+{
+    rb_thread_t *th = GET_THREAD();
+    rb_block_t block;
+    NODE *cref;
+    rb_proc_t *proc;
+
+    GetProcPtr(procval, proc);
+    block = proc->block;
+    block.self = self;
+    cref = vm_cref_push(th, under, NOEX_PUBLIC, &proc->block);
+    cref->flags |= NODE_FL_CREF_PUSHED_BY_EVAL;
+
+    return invoke_block_from_c(th, &block, self,
+	    RARRAY_LENINT(values), RARRAY_PTR(values),
+	    GC_GUARDED_PTR_REF(th->cfp->lfp[0]), cref);
+}
+
 /* string eval under the class/module context */
 static VALUE
 eval_under(VALUE under, VALUE self, VALUE src, const char *file, int line)

--
Yusuke Endoh [e-mail]

[mame (Yusuke Endoh)]

• Target version set to 2.6

[judofyr (Magnus Holm)]

I concur with trans: I don't think this issue will solve #6298. However, please keep that discussion out of this issue.

Would it be possible to get a OK/NG for inclusion of this in 2.1?

Still not sure about the name. Maybe #call_with is better?

[phluid61 (Matthew Kerwin)]

judofyr (Magnus Holm) wrote:

Still not sure about the name. Maybe #call_with is better?

I suggest #call_as or #call_bound: "as" because it suggests that "self" in the proc will refer to the first parameter, or possibly "bound" because we're binding the proc to the object (as well as to its existing binding scope). However I don't have strong feelings about any name, except that I like #call_with less than any others.

A side discussion: how does this tie in with currying? For example, do you foresee a Proc#curry_under (or curry_self, or curry_with, or whatever) method, that would cause the proc to be passed to instance_exec when all its parameters are satisfied?

[naruse (Yui NARUSE)]

• Target version deleted (2.6)

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)