Ruby » ruby-trunk

I'm curious, how you can be happy if you detect your VM supports tail-call optimization or not (from your script)?

I think you have to write without tail-calls anyway for compatibilities with those no-tail-call enironment.

Shyouhei Urabe schrieb:
> I'm curious, how you can be happy if you detect your VM supports tail-call optimization or not (from your script)?
> 
> I think you have to write without tail-calls anyway for compatibilities with those no-tail-call enironment.
I don't always write portable programs, but I want a controlled end of 
such a programm, e.g. with a message "Programm cannot run under this 
environment".

It's the same as I do in the moment. When started using Ruby 1.8 they 
tell the user, that it will not work using Ruby 1.8.

This s better than a system error message.

Otherwise it is also possible to present a different implementation, 
which may be much slower, but works and inform the user via a warning 
about this fact.

Wolfgang Nádasi-Donner

Hi,

At Mon, 9 Mar 2009 02:41:55 +0900,
Wolfgang Nádasi-Donner wrote in [ruby-core:22748]:
> Tail recursion optimization is an implementation detail and
> not a language feature in general. I propose to supply a
> constant TAILRECURSION, which will be set to "true" if this
> feature is implemented, otherwise "false".

RubyVM::InstructionSequence.compile_option[:tailcall_optimization]

-- 
Nobu Nakada

Nobuyoshi Nakada schrieb:
> Hi,
> 
> At Mon, 9 Mar 2009 02:41:55 +0900,
> Wolfgang Nádasi-Donner wrote in [ruby-core:22748]:
>> Tail recursion optimization is an implementation detail and
>> not a language feature in general. I propose to supply a
>> constant TAILRECURSION, which will be set to "true" if this
>> feature is implemented, otherwise "false".
> 
> RubyVM::InstructionSequence.compile_option[:tailcall_optimization]
> 
Thank you for this information. I saw that it still exists in Ruby 1.9.1-p0.

Yesterday I tried to compile Ruby 1.9.1-p0 with changes named by 
Nobuyoshi Nakada. In vm_opts.h I changed

#define OPT_TAILCALL_OPTIMIZATION    0

to

#define OPT_TAILCALL_OPTIMIZATION    1

Afterwards I rebuilt Ruby, but the stack overflow for both examples I 
posted before are still there.

Should I change additional things in the source?

Wolfgang Nádasi-Donner

As far as I understand this Ticket can be closed. If RubyVM::InstructionSequence.compile_option exists (defined? RubyVM::InstructionSequence.compile_option) and the value is "true" a program can expect a working tail recursion optimization, otherwise not.

Is this correct? - I yes, I think thee is no additional need for an extra constant.

Hi,

At Tue, 10 Mar 2009 04:26:51 +0900,
Wolfgang Nádasi-Donner wrote in [ruby-core:22785]:
> Afterwards I rebuilt Ruby, but the stack overflow for both examples I 
> posted before are still there.
> 
> Should I change additional things in the source?

trace_instruction option prevents the optimization.

$ cat ~/tmp/tailcall.rb
#! /usr/bin/ruby
src, file, line = <<SRC, __FILE__, __LINE__+1
def a(n)
  return if (n -= 1) <= 0
  a(n)
end
SRC
tailcallopt = ARGV[0] == "true"
traceinst = ARGV[1] != "false"
puts "tailcall_optimization: #{tailcallopt}, trace_instruction: #{traceinst}"
iseq = RubyVM::InstructionSequence.new(src, file, line,
                                       tailcall_optimization: tailcallopt,
                                       trace_instruction: traceinst)
iseq.eval
begin
  a(1000000)
  puts :ok
rescue SystemStackError => e
  puts "#{e.class} #{e.backtrace.size}"
end

$ ./ruby ~/tmp/tailcall.rb 
tailcall_optimization: false, trace_instruction: true
SystemStackError 8187

$ ./ruby ~/tmp/tailcall.rb true 
tailcall_optimization: true, trace_instruction: true
SystemStackError 8187

$ ./ruby ~/tmp/tailcall.rb true false
tailcall_optimization: true, trace_instruction: false
ok

-- 
Nobu Nakada