Routing problem with JRuby/Rails

[Daniel Hahn-3]

Hi everyone,

I just got started on JRuby, and tried to run our Rails application  
with it. I ran into a problem, which I debbugged (see description  
below) - however I'm not quite sure what to do with it. The root cause  
is rails assuming a particular quirk of the Dir[...] method, which  
JRuby doesn't replicate. So I have monkeypatched my Rails for now, but  
I'm not sure if this should be "fixed" in JRuby or taken to Rails...

Problem description:

JRuby interferes with routing where there is the following arrangement  
of controllers

     * something_controller.rb
     * something/ +- sub_controller.rb

this will not correctly match the 'something/sub' controller. The  
reason being:

     * The Route#recognize method will engage in creating meta-methods  
that match against regular expressions. (From Route#write_recognition)
     * For the default route, this will eventually land in  
ControllerSegment#regexp_chunk?
     * This will query Routing#possible_controllers
     * This method will collect the file paths from the controllers,  
using Dir.../**/*_controller.rb"

The controllers are then put into the regular expression.

For MRI, the Dir method reports files in subdirectories first. The  
JRuby implementation does not.

As a result the 'something' is put into the regexp first, and  
'something/sub' is matched as (something)/(sub). For MRI, 'something/
sub' is tested first, and it is correctly matched as (something/sub).

Ciao,
                 Daniel


--
Daniel Hahn
Software Architect
hahn@...

Net7
Via Marche 8a - 56123 Pisa
http://www.netseven.it/





---------------------------------------------------------------------
To unsubscribe from this list, please visit:

    http://xircles.codehaus.org/manage_email

[Nick Sieger-2]

On Wed, May 14, 2008 at 4:55 AM, Daniel Hahn <hahn@...> wrote:

> Problem description:
>
> JRuby interferes with routing where there is the following arrangement of
> controllers
>
>    * something_controller.rb
>    * something/ +- sub_controller.rb
>
> this will not correctly match the 'something/sub' controller. The reason
> being:
>
>    * The Route#recognize method will engage in creating meta-methods that
> match against regular expressions. (From Route#write_recognition)
>    * For the default route, this will eventually land in
> ControllerSegment#regexp_chunk?
>    * This will query Routing#possible_controllers
>    * This method will collect the file paths from the controllers, using
> Dir.../**/*_controller.rb"
>
> The controllers are then put into the regular expression.
>
> For MRI, the Dir method reports files in subdirectories first. The JRuby
> implementation does not.

Interesting! So you can confirm that the Dir[...] ordering is what
breaks it? We should be able to fix that -- I'd want to add to the
rubyspecs to ensure that this case is covered.

/Nick

---------------------------------------------------------------------
To unsubscribe from this list, please visit:

    http://xircles.codehaus.org/manage_email

[Vladimir Sizikov-2]

Hi Daniel,

Great analysis! Could you please submit a jira issue for that, with
some minimized example that shows the differences between MRI and
JRuby (otherwise, I'm not fully sure that I properly understand the
test case from your description).

Once we have the test, the fix should be (hopefully) simple. And, we
could put the test case  into rubyspecs too.

Thanks,
  --Vladimir

On Wed, May 14, 2008 at 11:55 AM, Daniel Hahn <hahn@...> wrote:

> Hi everyone,
>
> I just got started on JRuby, and tried to run our Rails application with it.
> I ran into a problem, which I debbugged (see description below) - however
> I'm not quite sure what to do with it. The root cause is rails assuming a
> particular quirk of the Dir[...] method, which JRuby doesn't replicate. So I
> have monkeypatched my Rails for now, but I'm not sure if this should be
> "fixed" in JRuby or taken to Rails...
>
> Problem description:
>
> JRuby interferes with routing where there is the following arrangement of
> controllers
>
>    * something_controller.rb
>    * something/ +- sub_controller.rb
>
> this will not correctly match the 'something/sub' controller. The reason
> being:
>
>    * The Route#recognize method will engage in creating meta-methods that
> match against regular expressions. (From Route#write_recognition)
>    * For the default route, this will eventually land in
> ControllerSegment#regexp_chunk?
>    * This will query Routing#possible_controllers
>    * This method will collect the file paths from the controllers, using
> Dir.../**/*_controller.rb"
>
> The controllers are then put into the regular expression.
>
> For MRI, the Dir method reports files in subdirectories first. The JRuby
> implementation does not.
>
> As a result the 'something' is put into the regexp first, and
> 'something/sub' is matched as (something)/(sub). For MRI, 'something/sub' is
> tested first, and it is correctly matched as (something/sub).
>
> Ciao,
>                Daniel
>
>
> --
> Daniel Hahn
> Software Architect
> hahn@...
>
> Net7
> Via Marche 8a - 56123 Pisa
> http://www.netseven.it/
>
>
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe from this list, please visit:
>
>   http://xircles.codehaus.org/manage_email
>
>
>

---------------------------------------------------------------------
To unsubscribe from this list, please visit:

    http://xircles.codehaus.org/manage_email

[Nick Sieger-2]

On Wed, May 14, 2008 at 8:32 AM, Vladimir Sizikov <vsizikov@...> wrote:
> Hi Daniel,
>
> Great analysis! Could you please submit a jira issue for that, with
> some minimized example that shows the differences between MRI and
> JRuby (otherwise, I'm not fully sure that I properly understand the
> test case from your description).
>
> Once we have the test, the fix should be (hopefully) simple. And, we
> could put the test case  into rubyspecs too.

This spec appears to cause the problem: http://is.gd/grR

I'll push it to the main rubyspec repo as soon as I can get a commit
bit from Brian.

/Nick

---------------------------------------------------------------------
To unsubscribe from this list, please visit:

    http://xircles.codehaus.org/manage_email

[Daniel Hahn-3]

Hi,

I was out of office yesterday afternoon. In any case, I don't have a  
test case but I can give a reasonably easy way to reproduce this. But  
is issue JRUBY-2519 (filed yesterday) already about this?

Ciao,
                Daniel


Il giorno 14/mag/08, alle ore 15:32, Vladimir Sizikov ha scritto:

> Hi Daniel,
>
> Great analysis! Could you please submit a jira issue for that, with
> some minimized example that shows the differences between MRI and
> JRuby (otherwise, I'm not fully sure that I properly understand the
> test case from your description).
>
> Once we have the test, the fix should be (hopefully) simple. And, we
> could put the test case  into rubyspecs too.
>
> Thanks,
>  --Vladimir
>
> On Wed, May 14, 2008 at 11:55 AM, Daniel Hahn <hahn@...>  
> wrote:
>> Hi everyone,
>>
>> I just got started on JRuby, and tried to run our Rails application  
>> with it.
>> I ran into a problem, which I debbugged (see description below) -  
>> however
>> I'm not quite sure what to do with it. The root cause is rails  
>> assuming a
>> particular quirk of the Dir[...] method, which JRuby doesn't  
>> replicate. So I
>> have monkeypatched my Rails for now, but I'm not sure if this  
>> should be
>> "fixed" in JRuby or taken to Rails...
>>
>> Problem description:
>>
>> JRuby interferes with routing where there is the following  
>> arrangement of
>> controllers
>>
>>   * something_controller.rb
>>   * something/ +- sub_controller.rb
>>
>> this will not correctly match the 'something/sub' controller. The  
>> reason
>> being:
>>
>>   * The Route#recognize method will engage in creating meta-methods  
>> that
>> match against regular expressions. (From Route#write_recognition)
>>   * For the default route, this will eventually land in
>> ControllerSegment#regexp_chunk?
>>   * This will query Routing#possible_controllers
>>   * This method will collect the file paths from the controllers,  
>> using
>> Dir.../**/*_controller.rb"
>>
>> The controllers are then put into the regular expression.
>>
>> For MRI, the Dir method reports files in subdirectories first. The  
>> JRuby
>> implementation does not.
>>
>> As a result the 'something' is put into the regexp first, and
>> 'something/sub' is matched as (something)/(sub). For MRI,  
>> 'something/sub' is
>> tested first, and it is correctly matched as (something/sub).
>>
>> Ciao,
>>               Daniel
>>
>>
>> --
>> Daniel Hahn
>> Software Architect
>> hahn@...
>>
>> Net7
>> Via Marche 8a - 56123 Pisa
>> http://www.netseven.it/
>>
>>
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe from this list, please visit:
>>
>>  http://xircles.codehaus.org/manage_email
>>
>>
>>
>
> ---------------------------------------------------------------------
> To unsubscribe from this list, please visit:
>
>    http://xircles.codehaus.org/manage_email
>
>

--
Daniel Hahn
Software Architect
hahn@...

Net7
Via Marche 8a - 56123 Pisa
http://www.netseven.it/





---------------------------------------------------------------------
To unsubscribe from this list, please visit:

    http://xircles.codehaus.org/manage_email

[Nick Sieger-2]

On Thu, May 15, 2008 at 4:59 AM, Daniel Hahn <hahn@...> wrote:
> Hi,
>
> I was out of office yesterday afternoon. In any case, I don't have a test
> case but I can give a reasonably easy way to reproduce this. But is issue
> JRUBY-2519 (filed yesterday) already about this?

No, I think this problem is suitably different and warrants its own
bug. I've filed one here: http://jira.codehaus.org/browse/JRUBY-2525

/Nick

---------------------------------------------------------------------
To unsubscribe from this list, please visit:

    http://xircles.codehaus.org/manage_email