Using gnutls.c by default

Using gnutls.c by default

[Lars Magne Ingebrigtsen]

I haven't seen any gnutls-related hangs in quite a while, so I tried by
trusty test case:

  (progn
   (require 'gnutls)
   (url-retrieve "https://www.google.no" #'ignore)
   (url-retrieve "https://www.google.no" #'ignore))

and it no longer hangs Emacs.

So I'm wondering whether `open-network-stream' should start using
gnutls.c by default now?  That would also give us greater test
coverage.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: Using gnutls.c by default

[Ted Zlatanov]

On Mon, 02 May 2011 15:54:03 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> I haven't seen any gnutls-related hangs in quite a while, so I tried by
LMI> trusty test case:

LMI>   (progn
LMI>    (require 'gnutls)
LMI>    (url-retrieve "https://www.google.no" #'ignore)
LMI>    (url-retrieve "https://www.google.no" #'ignore))

LMI> and it no longer hangs Emacs.

LMI> So I'm wondering whether `open-network-stream' should start using
LMI> gnutls.c by default now?  That would also give us greater test
LMI> coverage.

Juanma has some W32 improvements in the queue.  Also until the W32
builds have GnuTLS enabled I wouldn't make it the default.

Finally there are some minor issues left, and the documentation still
needs to be written, which I will do.

Ted

Re: Using gnutls.c by default

[Lars Magne Ingebrigtsen]

Ted Zlatanov <tzz@lifelogs.com> writes:

> Juanma has some W32 improvements in the queue.  Also until the W32
> builds have GnuTLS enabled I wouldn't make it the default.

Currently `open-network-stream' checks for `(fboundp
'open-gnutls-stream)', which means that it won't use that unless you've
loaded gnutls.el.  I was just thinking of checking for the C-level
`gnutls-book' instead, which would mean that it would be used by all
Emacsen that has gnutls.c compiled.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: Using gnutls.c by default

[Ted Zlatanov]

On Mon, 02 May 2011 20:11:27 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> Ted Zlatanov <tzz@lifelogs.com> writes:
>> Juanma has some W32 improvements in the queue.  Also until the W32
>> builds have GnuTLS enabled I wouldn't make it the default.

LMI> Currently `open-network-stream' checks for `(fboundp
LMI> 'open-gnutls-stream)', which means that it won't use that unless you've
LMI> loaded gnutls.el.  I was just thinking of checking for the C-level
LMI> `gnutls-book' instead, which would mean that it would be used by all
LMI> Emacsen that has gnutls.c compiled.

Juanma and I were just discussing having `gnutls-boot' throw an error if
the GnuTLS libraries can't be loaded.  If you can catch those errors in
`open-network-stream' even when `gnutls-boot' is defined, I think that's
even better, as it will DTRT in all the cases.

Ted

Re: Using gnutls.c by default

[Christoph Scholtes]

On 5/2/2011 11:57 AM, Ted Zlatanov wrote:

> Juanma has some W32 improvements in the queue.  Also until the W32
> builds have GnuTLS enabled I wouldn't make it the default.

Right now, if you enable GnuTLS support on Windows, you MUST have the 
GnuTLS libraries in order to even start Emacs. This is, I believe, what 
Juanma's improvements will fix (dynamic loading of the library if 
needed, much like the graphics libraries).

Until this is implemented, GnuTLS should not be enabled by default and 
Sean's weekly builds should also not be built with GnuTLS support enabled.

Christoph

Re: Using gnutls.c by default

[Juanma Barranquero]

On Tue, May 3, 2011 at 00:50, Christoph Scholtes
<cschol2112@googlemail.com> wrote:

> Right now, if you enable GnuTLS support on Windows, you MUST have the GnuTLS
> libraries in order to even start Emacs. This is, I believe, what Juanma's
> improvements will fix (dynamic loading of the library if needed, much like
> the graphics libraries).

Right.

> Until this is implemented, GnuTLS should not be enabled by default and
> Sean's weekly builds should also not be built with GnuTLS support enabled.

Agreed.

    Juanma

Re: Using gnutls.c by default

[Ted Zlatanov]

On Tue, 3 May 2011 01:09:02 +0200 Juanma Barranquero <lekktu@gmail.com> wrote: 

JB> On Tue, May 3, 2011 at 00:50, Christoph Scholtes
JB> <cschol2112@googlemail.com> wrote:

>> Right now, if you enable GnuTLS support on Windows, you MUST have the GnuTLS
>> libraries in order to even start Emacs. This is, I believe, what Juanma's
>> improvements will fix (dynamic loading of the library if needed, much like
>> the graphics libraries).

JB> Right.

>> Until this is implemented, GnuTLS should not be enabled by default and
>> Sean's weekly builds should also not be built with GnuTLS support enabled.

JB> Agreed.

Now that W32 has dynamic GnuTLS loading, I think it can be enabled by
default.  Does anyone have objections or comments?  As I mentioned I
don't use that platform so I can't speak for its users.

Ted

Re: Using gnutls.c by default

[Juanma Barranquero]

2011/5/5 Ted Zlatanov <tzz@lifelogs.com>:

> Now that W32 has dynamic GnuTLS loading, I think it can be enabled by
> default.  Does anyone have objections or comments?

Yes, definitely. I meant to send a message to Sean about this issue,
but I forgot.

    Juanma

Re: Using gnutls.c by default

[Andy Moreton]

On Thu 05 May 2011, Ted Zlatanov wrote:

> On Tue, 3 May 2011 01:09:02 +0200 Juanma Barranquero <lekktu@gmail.com> wrote: 
>
> JB> On Tue, May 3, 2011 at 00:50, Christoph Scholtes
> JB> <cschol2112@googlemail.com> wrote:
>
>>> Right now, if you enable GnuTLS support on Windows, you MUST have the GnuTLS
>>> libraries in order to even start Emacs. This is, I believe, what Juanma's
>>> improvements will fix (dynamic loading of the library if needed, much like
>>> the graphics libraries).
>
> JB> Right.
>
>>> Until this is implemented, GnuTLS should not be enabled by default and
>>> Sean's weekly builds should also not be built with GnuTLS support enabled.
>
> JB> Agreed.
>
> Now that W32 has dynamic GnuTLS loading, I think it can be enabled by
> default.  Does anyone have objections or comments?  As I mentioned I
> don't use that platform so I can't speak for its users.

There is little evidence of anybody using it yet, so making it the
default seems premature.

It would be better to wait a week or two until there has been some
feedback about use of GnuTLS on Win32.

    AndyM

Re: Using gnutls.c by default

[Juanma Barranquero]

On Thu, May 5, 2011 at 13:50, Andy Moreton <andrewjmoreton@gmail.com> wrote:

> It would be better to wait a week or two until there has been some
> feedback about use of GnuTLS on Win32.

We're talking about the trunk. Supporting GnuTLS by default is what
will give us feedback, at least on Windows, where (much) fewer people
is expected to build their own Emacs.

    Juanma

Re: Using gnutls.c by default

[Ted Zlatanov]

On Thu, 5 May 2011 13:57:48 +0200 Juanma Barranquero <lekktu@gmail.com> wrote: 

JB> On Thu, May 5, 2011 at 13:50, Andy Moreton <andrewjmoreton@gmail.com> wrote:
>> It would be better to wait a week or two until there has been some
>> feedback about use of GnuTLS on Win32.

JB> We're talking about the trunk. Supporting GnuTLS by default is what
JB> will give us feedback, at least on Windows, where (much) fewer people
JB> is expected to build their own Emacs.

So far it's 5 to 1 votes in favor; I think it's badly needed on W32
where there are serious problems with the alternatives to GnuTLS.

Note that the user still has to (require 'gnutls) to use it.  We're just
adding the library support to Emacs.  Without (require 'gnutls)
`open-network-stream' won't use `gnutls-negotiate'.

Ted

Re: Using gnutls.c by default

[Robert Pluim]

Ted Zlatanov <tzz@lifelogs.com> writes:

> On Thu, 5 May 2011 13:57:48 +0200 Juanma Barranquero <lekktu@gmail.com> wrote: 
>
> JB> On Thu, May 5, 2011 at 13:50, Andy Moreton <andrewjmoreton@gmail.com> wrote:
>>> It would be better to wait a week or two until there has been some
>>> feedback about use of GnuTLS on Win32.
>
> JB> We're talking about the trunk. Supporting GnuTLS by default is what
> JB> will give us feedback, at least on Windows, where (much) fewer people
> JB> is expected to build their own Emacs.
>
> So far it's 5 to 1 votes in favor; I think it's badly needed on W32
> where there are serious problems with the alternatives to GnuTLS.
>
> Note that the user still has to (require 'gnutls) to use it.  We're just
> adding the library support to Emacs.  Without (require 'gnutls)
> `open-network-stream' won't use `gnutls-negotiate'.

Oh, I thought Lars was proposing to go further than that, and always use
GnuTLS if available. Makes no difference to my opinion, GnuTLS works for
me :)

Robert

Re: Using gnutls.c by default

[Ted Zlatanov]

On Thu, 05 May 2011 16:16:44 +0200 Robert Pluim <rpluim@gmail.com> wrote: 

RP> Ted Zlatanov <tzz@lifelogs.com> writes:
>> On Thu, 5 May 2011 13:57:48 +0200 Juanma Barranquero <lekktu@gmail.com> wrote: 
>> 
JB> On Thu, May 5, 2011 at 13:50, Andy Moreton <andrewjmoreton@gmail.com> wrote:
>>>> It would be better to wait a week or two until there has been some
>>>> feedback about use of GnuTLS on Win32.
>> 
JB> We're talking about the trunk. Supporting GnuTLS by default is what
JB> will give us feedback, at least on Windows, where (much) fewer people
JB> is expected to build their own Emacs.
>> 
>> So far it's 5 to 1 votes in favor; I think it's badly needed on W32
>> where there are serious problems with the alternatives to GnuTLS.
>> 
>> Note that the user still has to (require 'gnutls) to use it.  We're just
>> adding the library support to Emacs.  Without (require 'gnutls)
>> `open-network-stream' won't use `gnutls-negotiate'.

RP> Oh, I thought Lars was proposing to go further than that, and always use
RP> GnuTLS if available. Makes no difference to my opinion, GnuTLS works for
RP> me :)

After re-reading the thread, you're right.  Sorry for the confusion, I
thought it was just about adding the library support.

So this proposal is to enable GnuTLS support for all the builds,
including W32, and then to make (require 'gnutls) unnecessary because it
will be done by `open-network-stream' by default.  The user has to
explicitly turn off gnutls.el support in order to get the old behavior.

Ted

Using gnutls.c by default

[T.V. Raman]

In that case, could we support slightly older versions of gnutls?
Apparently there is no prebuilt package for gnutls 2.6 on Jaunty
-- I'm going to install  it from source. But I suspect the
average user is going to get burnt.
-- 
Best Regards,
--raman

-- 
Best Regards,
--raman


On 5/5/11, Ted Zlatanov <tzz@lifelogs.com> wrote:
> On Thu, 05 May 2011 16:16:44 +0200 Robert Pluim <rpluim@gmail.com> wrote:
>
> RP> Ted Zlatanov <tzz@lifelogs.com> writes:
>>> On Thu, 5 May 2011 13:57:48 +0200 Juanma Barranquero <lekktu@gmail.com>
>>> wrote:
>>>
> JB> On Thu, May 5, 2011 at 13:50, Andy Moreton <andrewjmoreton@gmail.com>
> wrote:
>>>>> It would be better to wait a week or two until there has been some
>>>>> feedback about use of GnuTLS on Win32.
>>>
> JB> We're talking about the trunk. Supporting GnuTLS by default is what
> JB> will give us feedback, at least on Windows, where (much) fewer people
> JB> is expected to build their own Emacs.
>>>
>>> So far it's 5 to 1 votes in favor; I think it's badly needed on W32
>>> where there are serious problems with the alternatives to GnuTLS.
>>>
>>> Note that the user still has to (require 'gnutls) to use it.  We're just
>>> adding the library support to Emacs.  Without (require 'gnutls)
>>> `open-network-stream' won't use `gnutls-negotiate'.
>
> RP> Oh, I thought Lars was proposing to go further than that, and always use
> RP> GnuTLS if available. Makes no difference to my opinion, GnuTLS works for
> RP> me :)
>
> After re-reading the thread, you're right.  Sorry for the confusion, I
> thought it was just about adding the library support.
>
> So this proposal is to enable GnuTLS support for all the builds,
> including W32, and then to make (require 'gnutls) unnecessary because it
> will be done by `open-network-stream' by default.  The user has to
> explicitly turn off gnutls.el support in order to get the old behavior.
>
> Ted
>
>
>

Re: Using gnutls.c by default

[Ted Zlatanov]

On Sat, 7 May 2011 08:05:23 -0700 "T.V. Raman" <tv.raman.tv@gmail.com> wrote: 

TVR> In that case, could we support slightly older versions of gnutls?
TVR> Apparently there is no prebuilt package for gnutls 2.6 on Jaunty
TVR> -- I'm going to install  it from source. But I suspect the
TVR> average user is going to get burnt.

I would rather not support old versions of GnuTLS.  I believe we
discussed this a few months ago when we were talking about callbacks.
2.6 is already very old; remember that old versions have potential
security issues so we don't want to encourage their use.

The average user will get Emacs from a distribution, not compile it
themselves.  For those who compile, either they get a recent GnuTLS
distribution or they will have to specify --without-gnutls.  I think
that's reasonable.

If you disagree, you or someone else will have to support those older
versions.  The API is quite different so it's not trivial.

Ted

Re: Using gnutls.c by default

[Stefan Monnier]

> In that case, could we support slightly older versions of gnutls?
> Apparently there is no prebuilt package for gnutls 2.6 on Jaunty
> -- I'm going to install  it from source. But I suspect the
> average user is going to get burnt.

By the time Emacs-24 gets released this shouldn't be a big problem.


        Stefan

Re: Using gnutls.c by default

[Christoph Scholtes]

On 5/5/2011 4:42 AM, Ted Zlatanov wrote:

> Now that W32 has dynamic GnuTLS loading, I think it can be enabled by
> default.  Does anyone have objections or comments?  As I mentioned I
> don't use that platform so I can't speak for its users.

No objections. The dynamic loading works fine and Emacs starts up 
without the binaries present.

Thanks for adding this Juanma.

Christoph

Re: Using gnutls.c by default

[Juanma Barranquero]

On Thu, May 5, 2011 at 14:10, Christoph Scholtes
<cschol2112@googlemail.com> wrote:

> The dynamic loading works fine and Emacs starts up without
> the binaries present.

The real test is whether it crashes when the GnuTLS functions are invoked ;-)

> Thanks for adding this Juanma.

It's in the job description.

    Juanma

Re: Using gnutls.c by default

[Christoph Scholtes]

On 5/5/2011 6:15 AM, Juanma Barranquero wrote:

> The real test is whether it crashes when the GnuTLS functions are invoked ;-)

One step at a time. ;)

Re: Using gnutls.c by default

[Robert Pluim]

Ted Zlatanov <tzz@lifelogs.com> writes:

> On Tue, 3 May 2011 01:09:02 +0200 Juanma Barranquero <lekktu@gmail.com> wrote: 
>
> JB> On Tue, May 3, 2011 at 00:50, Christoph Scholtes
> JB> <cschol2112@googlemail.com> wrote:
>
>>> Right now, if you enable GnuTLS support on Windows, you MUST have the GnuTLS
>>> libraries in order to even start Emacs. This is, I believe, what Juanma's
>>> improvements will fix (dynamic loading of the library if needed, much like
>>> the graphics libraries).
>
> JB> Right.
>
>>> Until this is implemented, GnuTLS should not be enabled by default and
>>> Sean's weekly builds should also not be built with GnuTLS support enabled.
>
> JB> Agreed.
>
> Now that W32 has dynamic GnuTLS loading, I think it can be enabled by
> default.  Does anyone have objections or comments?  As I mentioned I
> don't use that platform so I can't speak for its users.

I've been using it under cygwin for a while now. I do still get hangs
occasionally, but they're definitely not GnuTLS' fault (my X server
hangs as well at the same time, and sometimes I hadn't even started
gnus, so gnutls.el would not have been loaded yet), and they're very
highly correlated with wifi + ip address changes. I'd say go for it.

Robert