Jython

[Henrik Eriksson]
> Does anyone have access to the socket and select code that Alan has
> produced so far? Or if Alan reads this first :-)

Hi Henrik.

No one asked for the code, so no-one has seen it.

I'll send it to you this evening.

> (I sent a private mail a few days ago).

Hmm, I didn't get that. My mail server is swamped with SPAM these
days, and I'm trying out a couple of new anti-SPAM tools. Perhaps your
message got lost in there.

> I'm very interested in getting my hands on the work that has been done
> and hopefully I can get some time to finish it off.

Well, to be completely honest, I think it would probably take you as
long to familiarise yourself with my code as it would to just read the
notes and try your own independent implementation. So you might be
better off to implement it yourself, using the notes as a guide.

I was thinking of suggesting it is a Summer of Code project as well: I
think it's the ideal size for such a project, meaty enough, but not
overly so.

Regards,

Alan.

On 3/6/07, Alan Kennedy <jython-dev@...> wrote:
> Well, to be completely honest, I think it would probably take you as
> long to familiarise yourself with my code as it would to just read the
> notes and try your own independent implementation. So you might be
> better off to implement it yourself, using the notes as a guide.

I've read your notes and will probably take a stab at implementing it
myself. It will probably take me a while though, since I don't have as
much time as I'd wish.

A starting question... The CherryPy3 server choked when it tried using
the current socket module. It turns out that it uses the object
returned from socket.socket.makefile(). In jython this is a
FileWrapper around a PyFile, but in the CPython module a
socket._fileobject is used. The problem was that CherryPy was looking
for a '\r\n' to end the header part of an http request. But from the
jython "fileobject" it only got a '\n'. Switching over to using
_fileobject worked fine. What was the reason for not using this in the
first place?

'H

PS: It would be great if you could send over the modified socket tests
you mention in a previous post.

[Henrik Eriksson]
>> Does anyone have access to the socket and select code that Alan has
>> produced so far? Or if Alan reads this first :-)

[Alan Kennedy]
> I'll send it to you this evening.

Well, would you believe it?

Yesterday evening, I went to tidy up that code, ready for handover.

Before I did that, I updated the JVM on my machine to JDK 1.4.2_13
(from the 1.4.2_09 it was before that).

And, lo and behold, when I ran the tests, they all passed!

And when I switch back to 1.4.2_09, the tests hang again.

Yay! Oh Serendipity! That's a result that I'm very happy with :-D

So, between the now-working support for non-blocking IO, and the basic
client-side SSL support I developed a couple of weeks ago, we've now
got socket support that is 100% up-to-date with cpython 2.5. And
working as well! Yay!

I still have some work to do on the UDP support, and on exception mapping.

So now the question becomes "How do we manage getting it into the
jython source base"?

I'm going to send a few followup emails about this.

Alan.

[Henrik Eriksson]
> A starting question... The CherryPy3 server choked when it tried using
> the current socket module. It turns out that it uses the object
> returned from socket.socket.makefile(). In jython this is a
> FileWrapper around a PyFile, but in the CPython module a
> socket._fileobject is used. The problem was that CherryPy was looking
> for a '\r\n' to end the header part of an http request. But from the
> jython "fileobject" it only got a '\n'. Switching over to using
> _fileobject worked fine. What was the reason for not using this in the
> first place?

Hmm, I'm not getting your natural language description here. Can you
please post code?

What I suspect is happening is that the (pseudo) file is being opened
in "Text" mode, i.e. like this

pseudofile = mysocket.makefile()

Note the implicit mode='r', which also implies text mode.

If you don't want the "\r\n"s to be translated to "\n"s, open the
pseudo-file in binary mode, like this

pseudofile = mysocket.makefile(mode="rb")

Does that solve the problem?

Alan.

This is awesome!  Is there something odd about the code that prevents
you from just uploading a patch to the tracker?  I'd love to get it
into 2.2

Charlie

On 3/7/07, Alan Kennedy <jython-dev@...> wrote:
> [Henrik Eriksson]
> >> Does anyone have access to the socket and select code that Alan has
> >> produced so far? Or if Alan reads this first :-)
>
> [Alan Kennedy]
> > I'll send it to you this evening.
>
> Well, would you believe it?
>
> Yesterday evening, I went to tidy up that code, ready for handover.
>
> Before I did that, I updated the JVM on my machine to JDK 1.4.2_13
> (from the 1.4.2_09 it was before that).
>
> And, lo and behold, when I ran the tests, they all passed!
>
> And when I switch back to 1.4.2_09, the tests hang again.
>
> Yay! Oh Serendipity! That's a result that I'm very happy with :-D
>
> So, between the now-working support for non-blocking IO, and the basic
> client-side SSL support I developed a couple of weeks ago, we've now
> got socket support that is 100% up-to-date with cpython 2.5. And
> working as well! Yay!
>
> I still have some work to do on the UDP support, and on exception mapping.
>
> So now the question becomes "How do we manage getting it into the
> jython source base"?
>
> I'm going to send a few followup emails about this.
>
> Alan.
>
> -------------------------------------------------------------------------
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to share your
> opinions on IT & business topics through brief surveys-and earn cash
> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
> _______________________________________________
> Jython-dev mailing list
> Jython-dev@...
> https://lists.sourceforge.net/lists/listinfo/jython-dev
>

[Charlie Groves]
> Is there something odd about the code that prevents
> you from just uploading a patch to the tracker?

Not particularly. There are 5 files involved

socket.py (rewritten from 2.1, now has 700+ lines instead of 361)
select.py (brand new module)
test_socket.py (ported from cpython 2.4)
test_select.py (ported from cpython 2.4)
test_socket_ssl.py (ported from cpython 2.4)

Although, as you'll see from followup messages, the code is really in
alpha status, and needs some solid field testing before being declared
fit for use.

Ideally, it should live in some sort of sandbox for a while, until
people do field testing. I know there are several maintainers of
cpython applications that would try to make them work on jython. For
instance, Pyro can currently only run the client on jython: the server
end relies on the select module.

> I'd love to get it into 2.2

And as you'll see from followup messages, I'd love to get it into 2.1!

Regards,

Alan.

On 3/7/07, Alan Kennedy <jython-dev@...> wrote:
> Hmm, I'm not getting your natural language description here. Can you
> please post code?
>
> What I suspect is happening is that the (pseudo) file is being opened
> in "Text" mode, i.e. like this
>
> pseudofile = mysocket.makefile()

You're probably right about what's happening. Cherrypy does this:

self.rfile = sock.makefile("r", self.rbufsize) #
cherrypy.wsgiserver.__init__ line 593

makefile in the "old" jython socket module (for _tcpsocket) looks like this:

    def makefile(self, mode="r", bufsize=-1):
        file = None
        if self.istream:
            if self.ostream:
                file = org.python.core.PyFile(self.istream, self.ostream,
                                              "<socket>", mode)
            else:
                file = org.python.core.PyFile(self.istream, "<socket>", mode)
        elif self.ostream:
            file = org.python.core.PyFile(self.ostream, "<socket>", mode)
        else:
            raise IOError, "both istream and ostream have been shut down"
        if file:
            return _tcpsocket.FileWrapper(self, file)

In CPython on the other hand it looks like this:

    def makefile(self, mode='r', bufsize=-1):
        return _fileobject(self._sock, mode, bufsize)

If I just replace the jython module with this one (and of course
include the _fileobject class) my particular use case works as
cherrypy expects it to do.

'H

On 3/7/07, Alan Kennedy <jython-dev@...> wrote:
> Ideally, it should live in some sort of sandbox for a while
The 2.3 branch is, in many  respects, a "sandbox".  I think it would
certainly be a fine place for this code.  Once shaken out a bit, we
could back port it to 2.2.

-Frank

[Alan Kennedy]
>> Ideally, it should live in some sort of sandbox for a while

[Frank Wierzbicki]
> The 2.3 branch is, in many  respects, a "sandbox".  I think it would
> certainly be a fine place for this code.  Once shaken out a bit, we
> could back port it to 2.2.

Thanks Frank, that sounds like a good plan.

How best to proceed?

Should I just stop beating around the bush and request SVN commit priviledges?

Or should make it available as a patch and let someone else do the commit?

Regards,

Alan.