Resetting ruby


#1

I’m looking for a fast way to “reset” Ruby. That is, to reset the
interpreter back to a fresh state, or, even better, a fresh state +
keeping some required ruby-gems.

My goal is to make a continuos test runner, and avoid the overhead of
having to restart ruby and require the gems each times.

I could do this with fork (copy-on-write), except I want this to run on
Win32 as well.

Any ideas?


#2

On Jan 15, 2006, at 8:29 PM, List R. wrote:

Any ideas?


Posted via http://www.ruby-forum.com/.

I’m confused, why do you need to “reset” ruby everytime? Can’t you
stick the tests in a loop?


#3

Logan C. wrote:

I’m confused, why do you need to “reset” ruby everytime? Can’t you
stick the tests in a loop?

The code - and it’s require files - are constantly changing. (That’s
the whole point of continuos testing!). I want to a) make sure to
reload all required files and b) make sure to start with a fresh slate,
in terms of class definitions, ObjectSpace, and the like.


#4

On Jan 15, 2006, at 8:47 PM, List R. wrote:


Posted via http://www.ruby-forum.com/.

Ok, so change the 'require’s to 'load’s and then stick it in a
loop ;). In seriousness though, all the work involved with undefining
everything, and you might as well just start a new ruby executable.
Perhaps I’m not understanding you correctly, how would you do this
with fork, and avoid the overhead of requiring things?


#5

On 1/15/06, List R. removed_email_address@domain.invalid wrote:

Logan C. wrote:

Ok, so change the 'require’s to 'load’s and then stick it in a
loop ;). In seriousness though, all the work involved with undefining
everything, and you might as well just start a new ruby executable.
Perhaps I’m not understanding you correctly, how would you do this
with fork, and avoid the overhead of requiring things?

fork would save the overhead of loading ruby itself

However, with fork you’d need to be careful about open filehandles and
the like which are shared between the processes. Getting consistent
results between a stay alive process like this and a new process each
time can be really tricky.

Jacob F.


#6

On 1/15/06, Logan C. removed_email_address@domain.invalid wrote:

On Jan 15, 2006, at 8:47 PM, List R. wrote:

The code - and it’s require files - are constantly changing. (That’s
the whole point of continuos testing!). I want to a) make sure to
reload all required files and b) make sure to start with a fresh
slate,
in terms of class definitions, ObjectSpace, and the like.

Ok, so change the 'require’s to 'load’s and then stick it in a
loop ;).

That’s not quite sufficient. ‘load’ simply loads the file again. It
doesn’t replace the result of previously loading the file. So if the
“signature” of the file is the same (all the same classes, methods,
variables, etc. declared only with different values/implementations)
you’re good. But in the more frequent case where the signature changes
in a non-additive way (e.g. a method is removed), that change is not
propagated by a new load. That means the method that was removed is
still available within the test script, which may cause some tests to
pass which should fail and would fail if run under a new instance.

Jacob F.


#7

On Jan 16, 2006, at 12:39 AM, Jacob F. wrote:

in a non-additive way (e.g. a method is removed), that change is not
propagated by a new load. That means the method that was removed is
still available within the test script, which may cause some tests to
pass which should fail and would fail if run under a new instance.

Jacob F.

I know, hence the smiley face. I still think this idea is a little
extreme for the minor savings you’d get out of it.


#8

Logan C. wrote:

On Jan 15, 2006, at 8:47 PM, List R. wrote:


Posted via http://www.ruby-forum.com/.

Ok, so change the 'require’s to 'load’s and then stick it in a
loop ;). In seriousness though, all the work involved with undefining
everything, and you might as well just start a new ruby executable.
Perhaps I’m not understanding you correctly, how would you do this
with fork, and avoid the overhead of requiring things?

fork would save the overhead of loading ruby itself


#9

On 1/16/06, Dalibor S. removed_email_address@domain.invalid wrote:

On Mon, Jan 16, 2006 at 02:45:46PM +0900, Logan C. wrote:

I know, hence the smiley face. I still think this idea is a little
extreme for the minor savings you’d get out of it.

The possibility to reset the interpreter would be IMHO useful also for other
situations where you have persistent Ruby processes (think mod_ruby).

It would also be useful for debuggers so you could rerun the program
without having to re-enter breakpoints. As far as I know, this isn’t
currently possible with the built-in Ruby debugger. Can any Ruby
debuggers do this now?


#10

On 1/16/06, List R. removed_email_address@domain.invalid wrote:

I’m looking for a fast way to “reset” Ruby. That is, to reset the
interpreter back to a fresh state, or, even better, a fresh state +
keeping some required ruby-gems.

My goal is to make a continuos test runner, and avoid the overhead of
having to restart ruby and require the gems each times.

I’ve got code for this kind of testrunner; no specific optimizations,
though. It’s a plain testrunner using pipes for communication and a
fork to separate out the execution. There’s something clumsy about it

  • I don’t remember what it was.

If you’re dead set on doing this inside a single Ruby process, this
won’t help - if you want an easy way out, I’ll try to dig up the code

  • it wasn’t in the obvious places :-/

Eivind.


#11

On Mon, Jan 16, 2006 at 02:45:46PM +0900, Logan C. wrote:

I know, hence the smiley face. I still think this idea is a little
extreme for the minor savings you’d get out of it.

The possibility to reset the interpreter would be IMHO useful also for
other
situations where you have persistent Ruby processes (think mod_ruby).

Dalibor S.


#12

List R. wrote:

I’m looking for a fast way to “reset” Ruby. That is, to reset the
interpreter back to a fresh state, or, even better, a fresh state +
keeping some required ruby-gems.

My goal is to make a continuous test runner, and avoid the overhead
of
having to restart ruby and require the gems each times.

One thing you could do would be to load up all of your libraries and
pause on the terminal waiting for user input. As soon as the user
unpaused the process it would load in the test cases and run them. As
the user was looking at the output and fixing tests you could load up
another instance and pause waiting for user input. Stir and repeat…

I’ve thought for a long time that something like this would be very
useful for Rails testing where loading up the library code takes 2-3
seconds, making unit testing tedious.


#13

Dalibor S. wrote:

The possibility to reset the interpreter would be IMHO useful also for other
situations where you have persistent Ruby processes (think mod_ruby).

John W. Long wrote:

I’ve thought for a long time that something like this would be very
useful for Rails testing where loading up the library code takes 2-3
seconds, making unit testing tedious.

Okay, Ruby core hackers, can you answer our call?!


#14

On 16 janv. 06, at 12:33, Dalibor S. wrote:

The possibility to reset the interpreter would be IMHO useful also
for other
situations where you have persistent Ruby processes (think mod_ruby).

(going slightly offtopic)

I have recently dived into Tcl (the “Practical Programming in Tcl and
Tk” book by Welch, Jones and Hobbs is pretty good) and while I still
haven’t managed to go past the “o-m-f-g-YUCK!” state as far as the
core language is concerned, I must say that the core implementation
provides some really nifty features which I would really love to
see in Ruby and which would incidentally make the original poster
question trivial to do.

For example, you can spawn clean and fresh “sub-interpreter(s)” which
you can then use as isolated (and optionally safe) sandboxe(s).
Here’s a sample usage:

% interp create worker
worker
% interp eval worker {set a 42}
42
% puts $a
can’t read “a”: no such variable
% interp delete worker

In the first line we create a sub-interpreter and name it ‘worker’.
In the second line we execute a simple command: we set assign the
value 42 to the variable ‘a’. In the third line we try to print the
value of ‘a’, which causes an error. This shows that the ‘main’
interpreter and our ‘worker’ sub-interpreter are clearly isolated
from each other. Finally, we get rid of the sub-interpreter.

Another (different) aspect of Tcl which I really like is how event
loops are used to provide ultra-simple interthreads communication
with message queuing and all that jazz:

% package require Thread
% set worker [thread::create]
% thread::send -async $worker {…some longish calculation…}
% thread::send -async $worker {…some other longish calculation…}
% …continue on the main thread…

Rite is supposed to be reentrant, so I guess that it shouldn’t be
very hard to build such things upon Ruby 2.0. Oh my, I am so
looking forward to it… :slight_smile:


#15

Luc H. wrote:

Rite is supposed to be reentrant, so I guess that it shouldn’t be very
hard to build such things upon Ruby 2.0. Oh my, I am so looking
forward to it… :slight_smile:

Aside from the use of native threads in this tcl example (IIRC), this is
not hard to do in ruby. And with 2.0, we’ll have native threads…

Anyway here’s one way to do worker threads:

require ‘thread’

class WorkerThread < Thread
def initialize
super do
@queue = Queue.new
loop do
@queue.pop.call
end
end
end

def push(&bl)
@queue << bl
end
end

def very_long_calculation; (10**10000).to_s.size; end

qt = WorkerThread.new
results = Queue.new

qt.push do
results.push(very_long_calculation)
end

p results.pop


#16

List R. wrote:

Okay, Ruby core hackers, can you answer our call?!

I don’t think there’s a need for Ruby core hacking if you follow my
suggestion:

One thing you could do would be to load up all of your libraries and
pause on the terminal waiting for user input. As soon as the user
unpaused the process it would load in the test cases and run them.
As

the user was looking at the output and fixing tests you could load
up

another instance and pause waiting for user input. Stir and
repeat…


#17

John W. Long wrote:

unpaused the process it would load in the test cases and run them. As
the user was looking at the output and fixing tests you could load up
another instance and pause waiting for user input. Stir and repeat…

That would work in this one case where resetting is useful. As this
thread has shown, there are others. Here is mine: I want to use an
irb-like environment in conjunction with gsl as a kind of super
calculator/numerical test engine. A reset method would be tremendously
useful as one often wants to start with a clean slate, without losing
all the output from the previous computations.

So, I second List R.'s plea to the core hackers: ideas anyone?


#18

Michael U. wrote:

One thing you could do would be to load up all of your libraries and
useful as one often wants to start with a clean slate, without losing
all the output from the previous computations.

So, I second List R.'s plea to the core hackers: ideas anyone?

Doesn’t GC clean the slate well enough for calculations?


#19

On Jan 15, 2006, at 5:29 PM, List R. wrote:

I’m looking for a fast way to “reset” Ruby. That is, to reset the
interpreter back to a fresh state, or, even better, a fresh state +
keeping some required ruby-gems.

My goal is to make a continuos test runner, and avoid the overhead of
having to restart ruby and require the gems each times.

In order to ensure your tests won’t fail or pass when they should do
the opposite you should start up from a clean interpreter.


Eric H. - removed_email_address@domain.invalid - http://segment7.net
This implementation is HODEL-HASH-9600 compliant

http://trackmap.robotcoop.com