You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Eric Gillespie <ep...@pretzelnet.org> on 2007/06/08 21:04:11 UTC

Serf as default DAV client?

Whatever happened to making serf the default DAV layer in 1.5?
Last i heard, we were going to follow fsfs in this respect.  fsfs
was an option in 1.1 and the default in 1.2.  Serf was an option
in 1.4, shouldn't we make it the default now?

If we don't switch to serf, it's never going to get the kind of
attention it needs...

-- 
Eric Gillespie <*> epg@pretzelnet.org

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by David Glasser <gl...@mit.edu>.
On 6/11/07, Daniel Rall <dl...@collab.net> wrote:
> Looking at it another way, it's not a whole lot different than adding
> a new parameter to a rev'd API, and hard-coding a default value for
> the older incarnation of the API.  So long as we document the
> behavior, I'd say we're okay.

OK.  Should it be documented in the source, or actually in the default
servers file?


Anyway, here is a preliminary patch to support building both ra_dav
and ra_serf and choosing which one in the servers file:

Some issues:

  * The logic is hardcoded around assuming that the only conflicts are
    ra_dav and ra_serf.  I think that's fine for now.

  * It only checks the [global] section, not the host-specific
    sections.  This is probably not acceptable (if we want to not have
    host-specific preferences, then the pref should be in the main
    config file, not the servers file).  All the code to choose
    sections is in libsvn_ra_dav, though.

  * The error thrown if the http-library is unknown is in the
    SVN_ERR_RA_DAV namespace.  It's not entirely inappropriate though.

  * The http-library value is validated on any use of any RA library,
    even a non-http[s] one.  I think that's OK though; nobody is going
    to see that error unless they *just* tweaked the value.

  * The fact that http-library is ignored by svn_ra_get_ra_library is
    not documented anywhere.

[[[
Support building both ra_dav and ra_serf, allowing the user to choose
which one to use in their servers config file.

* build/ac-macros/serf.m4:
  Remove check that prevents building both ra_dav and ra_serf.

* subversion/include/svn_config.h
  (SVN_CONFIG_OPTION_HTTP_LIBRARY): New config option "http-library".

* subversion/libsvn_subr/config_file.c
  (svn_config_ensure): Document SVN_CONFIG_OPTION_HTTP_LIBRARY in the
   generated servers file.

* subversion/libsvn_ra/ra_loader.c
  (): Change comment about conflicting RA libraries to mention the
   assumptions that svn_ra_open2 is making.
  (MUST_CHOOSE_DAV): Define this symbol if we are linking both DAV
   libraries.
  (svn_ra_open2): If MUST_CHOOSE_DAV is defined, look up the
   http-library value in the servers config file global section,
   validate that it is "dav" or "serf", and skip the other library.
]]]

-- 
David Glasser | glasser@mit.edu | http://www.davidglasser.net/

Re: Serf as default DAV client?

Posted by Daniel Rall <dl...@collab.net>.
On Sat, 09 Jun 2007, David Glasser wrote:

> On 6/8/07, Daniel Rall <dl...@collab.net> wrote:
> >I believe it's just a side-effect of the current RA loader
> >implementation.  IIRC, people have looked at changing this a few times
> >in the past, but didn't get any traction.  The RA loader isn't a ton
> >of code -- it should be pretty easy to just toss the existing one an
> >re-write it from scratch, if need be.
> 
> One problem with this.  RA libraries are chosen by two APIs:
> svn_ra_open[2] and svn_ra_get_ra_library.  svn_ra_open[2] gets a
> config hash, so it could use the servers file to decide which library
> to use.  But the compatibility interface svn_ra_get_ra_library does
> not get it.  If a program which uses svn_ra_get_ra_library (using the
> pre-1.2 API) is run against a Subversion client linking both ra_dav
> and ra_serf, I don't see any way for the config files to affect which
> is chosen, so even though your config file might say "use serf!", the
> internal fact that the program you're running uses the pre-1.2 API
> will prevent your configuration from working.
> 
> How big of an issue is this?  Is it acceptable to implement
> user-configurable RA library support and have it be ignored for
> programs written against the pre-1.2 API?  (Note that a user of such a
> program may have no idea what APIs it uses...)

IMO, the value of the configurability outweighs the minor
inconsistency in the API.

Looking at it another way, it's not a whole lot different than adding
a new parameter to a rev'd API, and hard-coding a default value for
the older incarnation of the API.  So long as we document the
behavior, I'd say we're okay.

Re: Serf as default DAV client?

Posted by David Glasser <gl...@mit.edu>.
On 6/8/07, Daniel Rall <dl...@collab.net> wrote:
> I believe it's just a side-effect of the current RA loader
> implementation.  IIRC, people have looked at changing this a few times
> in the past, but didn't get any traction.  The RA loader isn't a ton
> of code -- it should be pretty easy to just toss the existing one an
> re-write it from scratch, if need be.

One problem with this.  RA libraries are chosen by two APIs:
svn_ra_open[2] and svn_ra_get_ra_library.  svn_ra_open[2] gets a
config hash, so it could use the servers file to decide which library
to use.  But the compatibility interface svn_ra_get_ra_library does
not get it.  If a program which uses svn_ra_get_ra_library (using the
pre-1.2 API) is run against a Subversion client linking both ra_dav
and ra_serf, I don't see any way for the config files to affect which
is chosen, so even though your config file might say "use serf!", the
internal fact that the program you're running uses the pre-1.2 API
will prevent your configuration from working.

How big of an issue is this?  Is it acceptable to implement
user-configurable RA library support and have it be ignored for
programs written against the pre-1.2 API?  (Note that a user of such a
program may have no idea what APIs it uses...)

--dave

-- 
David Glasser | glasser@mit.edu | http://www.davidglasser.net/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On 6/8/07, Daniel Rall <dl...@collab.net> wrote:
> I believe it's just a side-effect of the current RA loader
> implementation.  IIRC, people have looked at changing this a few times
> in the past, but didn't get any traction.  The RA loader isn't a ton
> of code -- it should be pretty easy to just toss the existing one an
> re-write it from scratch, if need be.

Sure - it just needs someone to bother to rewrite it.  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Daniel Rall <dl...@collab.net>.
On Fri, 08 Jun 2007, David Glasser wrote:

> On 6/8/07, Daniel Rall <dl...@collab.net> wrote:
> >Because both RA layers handle DAV access, they currently can't
> >co-exist in the same client.
> 
> Is this just the result of the current implementation of ra_loader, or
> is this a policy decision for a good reason?  I mean, it's not like we
> don't already have a per-server config file.

I believe it's just a side-effect of the current RA loader
implementation.  IIRC, people have looked at changing this a few times
in the past, but didn't get any traction.  The RA loader isn't a ton
of code -- it should be pretty easy to just toss the existing one an
re-write it from scratch, if need be.

Re: Serf as default DAV client?

Posted by David Glasser <gl...@mit.edu>.
On 6/8/07, Daniel Rall <dl...@collab.net> wrote:
> Because both RA layers handle DAV access, they currently can't
> co-exist in the same client.

Is this just the result of the current implementation of ra_loader, or
is this a policy decision for a good reason?  I mean, it's not like we
don't already have a per-server config file.

--dave

-- 
David Glasser | glasser@mit.edu | http://www.davidglasser.net/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Eric Gillespie <ep...@pretzelnet.org>.
Daniel Rall <dl...@collab.net> writes:

> > How so?  No one's proposing removing neon.
> 
> Because both RA layers handle DAV access, they currently can't
> co-exist in the same client.  The *default* build should support SSPI.

Well, we don't distribute binaries, so you can build with
whichever implementation you want.  I'm not sure why --with-neon
is such a burden to someone capable of operating configure/make
machinery.  But if that's the consensus, it's fine with me ;->.
It just means i need to factor "neon is where it's at for the
foreseeable future" into my plans.

--  
Eric Gillespie <*> epg@pretzelnet.org

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Daniel Rall <dl...@collab.net>.
On Fri, 08 Jun 2007, Eric Gillespie wrote:

> Daniel Rall <dl...@collab.net> writes:
> 
> > Doesn't Serf still lack the SSPI/GSSAPI support present in Neon?
> > Dropping this functionality in our default DAV client would be a
> > serious regression.
> 
> How so?  No one's proposing removing neon.

Because both RA layers handle DAV access, they currently can't
co-exist in the same client.  The *default* build should support SSPI.

Re: Serf as default DAV client?

Posted by Eric Gillespie <ep...@pretzelnet.org>.
Daniel Rall <dl...@collab.net> writes:

> Doesn't Serf still lack the SSPI/GSSAPI support present in Neon?
> Dropping this functionality in our default DAV client would be a
> serious regression.

How so?  No one's proposing removing neon.

--  
Eric Gillespie <*> epg@pretzelnet.org

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Daniel Rall <dl...@collab.net>.
On Fri, 08 Jun 2007, Eric Gillespie wrote:

> Whatever happened to making serf the default DAV layer in 1.5?
> Last i heard, we were going to follow fsfs in this respect.  fsfs
> was an option in 1.1 and the default in 1.2.  Serf was an option
> in 1.4, shouldn't we make it the default now?

Doesn't Serf still lack the SSPI/GSSAPI support present in Neon?
Dropping this functionality in our default DAV client would be a
serious regression.

Re: Serf as default DAV client?

Posted by Stefan Küng <to...@gmail.com>.
David Glasser wrote:
> On 6/8/07, Eric Gillespie <ep...@pretzelnet.org> wrote:
>> Whatever happened to making serf the default DAV layer in 1.5?
>> Last i heard, we were going to follow fsfs in this respect.  fsfs
>> was an option in 1.1 and the default in 1.2.  Serf was an option
>> in 1.4, shouldn't we make it the default now?
>>
>> If we don't switch to serf, it's never going to get the kind of
>> attention it needs...
> 
> Are many people running it as their day-to-day implementation?  I
> recently tried to run tests for a patch I was working on over serf,
> and found several places where it handled errors (like invalid user
> input or network problems, not internal corruption) with aborts... I
> fixed one or two of those places, but it definitely didn't fill me
> with confidence that ra_serf, in its current incarnation, is ready to
> be our primary implementation.  (In my copious free time I'd like to
> learn how serf itself works so I can be more help in fixing these
> issues...)

Well, the abort() calls should be gone now, at least the issue I opened 
about it got closed:
http://code.google.com/p/serf/issues/detail?id=13

But there are still issues left before serf can really replace neon (at 
least for me):
http://code.google.com/p/serf/issues/list

Stefan

-- 
        ___
   oo  // \\      "De Chelonian Mobile"
  (_,\/ \_/ \     TortoiseSVN
    \ \_/_\_/>    The coolest Interface to (Sub)Version Control
    /_/   \_\     http://tortoisesvn.net

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Ben Collins-Sussman <su...@red-bean.com>.
Woohoo, I just rebuilt serf, and indeed, it works against google code now.

I'm gonna leave serf in my build for now, so it gets more testing...


On 6/11/07, Justin Erenkrantz <ju...@erenkrantz.com> wrote:
> On 6/11/07, Justin Erenkrantz <ju...@erenkrantz.com> wrote:
> > Well, to be quite emphatic: Google does *not* support HTTP/1.1 - it is
> > a quite blatant RFC violation on Google's part.  While I was at
>
> Ben poked me off-list to say that the internal bug on Google's side is
> now marked as fixed - and I can confirm that.  I've now got serf
> working against Google Code in serf r1111.  Google is aggressive with
> closing connections without telling us - that's okay - so I taught
> serf in r1111 to play nice with that behavior.
>
> And there was much rejoicing!  Thanks!  -- justin
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: dev-help@subversion.tigris.org
>
>

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On 6/11/07, Justin Erenkrantz <ju...@erenkrantz.com> wrote:
> Well, to be quite emphatic: Google does *not* support HTTP/1.1 - it is
> a quite blatant RFC violation on Google's part.  While I was at

Ben poked me off-list to say that the internal bug on Google's side is
now marked as fixed - and I can confirm that.  I've now got serf
working against Google Code in serf r1111.  Google is aggressive with
closing connections without telling us - that's okay - so I taught
serf in r1111 to play nice with that behavior.

And there was much rejoicing!  Thanks!  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Erik Huelsmann <eh...@gmail.com>.
On 6/12/07, John Szakmeister <jo...@szakmeister.net> wrote:
> Justin Erenkrantz wrote:
> > On 6/11/07, Dan Christian <dc...@google.com> wrote:
> >> I've got 2 major operational problems with serf at the moment (this is
> >> after
> >> Jason fixing several issues.  Thanks!).  Neither is technically a serf
> >> issue, but they none the less make serf un-usable.
> >>
> >> serf over SSL trips a core dump due to a double free bug in
> >> libcrypto-0.9.8/libssl-0.9.8.  neon hits this as well, but serf hits
> >> it much
> >> more often.  I can't get through any of my big test cases without a
> >> fault.
> >
> > Huh.  If you can provide some information about this, I'll see what we
> > can do - but if Neon hits it too, then it's likely an OpenSSL bug...
> > =(
> >
> >> serf doesn't play well with some http-1.1 implementations (specifically
> >> code.google.com).  neon does fine here.  It raises the issue of how
> >> many bad
> >> http-1.1 implementations there are and how to validate serf to be "good
> >> enough" to be the default.
> >
> > Well, to be quite emphatic: Google does *not* support HTTP/1.1 - it is
> > a quite blatant RFC violation on Google's part.  While I was at
> > Google, Greg and I talked at length about whether we should work
> > around it in serf - and we came to the opinion that it is reasonable
> > for serf to expect RFC-compliant server implementations.  Everyone
> > else is going to be using Apache or IIS or Squid which doesn't have
> > this particular brain damage with respect to the RFC - and,
> > realistically, almost everyone will be using Apache 2.x.  So, I don't
> > support crippling serf just to support non-compliant HTTP/1.1 server
> > implementations - even if it is Google.  -- justin
>
> I understand the desire for purity... but if svn all of a sudden became
> incompatible while Google Code, I believe people would see that as a
> regression.  I can already see the flame wars on users@...
>
> Just my 2 cents.

Hi John,

I was about to say the same, but the whole point has turned moot now
that Google Code has been fixed and Serf has been updated to support
Google's interpretation of the RFC (see message by Justin about that).

Bye,

Erik.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by John Szakmeister <jo...@szakmeister.net>.
Justin Erenkrantz wrote:
> On 6/11/07, Dan Christian <dc...@google.com> wrote:
>> I've got 2 major operational problems with serf at the moment (this is 
>> after
>> Jason fixing several issues.  Thanks!).  Neither is technically a serf
>> issue, but they none the less make serf un-usable.
>>
>> serf over SSL trips a core dump due to a double free bug in
>> libcrypto-0.9.8/libssl-0.9.8.  neon hits this as well, but serf hits 
>> it much
>> more often.  I can't get through any of my big test cases without a 
>> fault.
> 
> Huh.  If you can provide some information about this, I'll see what we
> can do - but if Neon hits it too, then it's likely an OpenSSL bug...
> =(
> 
>> serf doesn't play well with some http-1.1 implementations (specifically
>> code.google.com).  neon does fine here.  It raises the issue of how 
>> many bad
>> http-1.1 implementations there are and how to validate serf to be "good
>> enough" to be the default.
> 
> Well, to be quite emphatic: Google does *not* support HTTP/1.1 - it is
> a quite blatant RFC violation on Google's part.  While I was at
> Google, Greg and I talked at length about whether we should work
> around it in serf - and we came to the opinion that it is reasonable
> for serf to expect RFC-compliant server implementations.  Everyone
> else is going to be using Apache or IIS or Squid which doesn't have
> this particular brain damage with respect to the RFC - and,
> realistically, almost everyone will be using Apache 2.x.  So, I don't
> support crippling serf just to support non-compliant HTTP/1.1 server
> implementations - even if it is Google.  -- justin

I understand the desire for purity... but if svn all of a sudden became 
incompatible while Google Code, I believe people would see that as a 
regression.  I can already see the flame wars on users@...

Just my 2 cents.

-John

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On 6/11/07, Dan Christian <dc...@google.com> wrote:
> I've got 2 major operational problems with serf at the moment (this is after
> Jason fixing several issues.  Thanks!).  Neither is technically a serf
> issue, but they none the less make serf un-usable.
>
> serf over SSL trips a core dump due to a double free bug in
> libcrypto-0.9.8/libssl-0.9.8.  neon hits this as well, but serf hits it much
> more often.  I can't get through any of my big test cases without a fault.

Huh.  If you can provide some information about this, I'll see what we
can do - but if Neon hits it too, then it's likely an OpenSSL bug...
=(

> serf doesn't play well with some http-1.1 implementations (specifically
> code.google.com).  neon does fine here.  It raises the issue of how many bad
> http-1.1 implementations there are and how to validate serf to be "good
> enough" to be the default.

Well, to be quite emphatic: Google does *not* support HTTP/1.1 - it is
a quite blatant RFC violation on Google's part.  While I was at
Google, Greg and I talked at length about whether we should work
around it in serf - and we came to the opinion that it is reasonable
for serf to expect RFC-compliant server implementations.  Everyone
else is going to be using Apache or IIS or Squid which doesn't have
this particular brain damage with respect to the RFC - and,
realistically, almost everyone will be using Apache 2.x.  So, I don't
support crippling serf just to support non-compliant HTTP/1.1 server
implementations - even if it is Google.  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Dan Christian <dc...@google.com>.
I've got 2 major operational problems with serf at the moment (this is after
Jason fixing several issues.  Thanks!).  Neither is technically a serf
issue, but they none the less make serf un-usable.

serf over SSL trips a core dump due to a double free bug in libcrypto-0.9.8
/libssl-0.9.8.  neon hits this as well, but serf hits it much more often.  I
can't get through any of my big test cases without a fault.

serf doesn't play well with some http-1.1 implementations (specifically
code.google.com).  neon does fine here.  It raises the issue of how many bad
http-1.1 implementations there are and how to validate serf to be "good
enough" to be the default.

-Dan C

On 6/8/07, Justin Erenkrantz <ju...@erenkrantz.com> wrote:
>
> On 6/8/07, Eric Gillespie <ep...@pretzelnet.org> wrote:
> > No, we've tried at Google, as well.  It's got a long way to go.
> > That's my point: as long as it sits there unused, it's not going
> > to get any better.  As a matter of fact, it seems likely to me
> > that it's only going to get worse.
>
> I kinda doubt it's got a 'long way to go' - it passes all of the
> regression tests on trunk - or did the last time I checked.  True,
> ra_serf may need some polishing or a feature here or there - but I
> don't think it's going to need *massive* amounts of work - just people
> who are willing to add the features they think are missing before it
> is acceptable to them.  But, it shouldn't need a tremendous amount of
> effort as all of the hard stuff is already done...
>
> FWIW, on serf-dev, Lieven said that he was looking into SSPI and that
> it looked fairly straightforward to add.  *shrug*
>
> Again, ra_serf 'works for me' - I use it without issues.  I'd advocate
> flipping the default on trunk and any lingering bugs will get worked
> out really fast.  =)  For those who really want the features that only
> neon provides, it'll still be there ready to be used - I don't see the
> harm in trying it.  -- justin
>
>

Re: Serf as default DAV client?

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On 6/8/07, Eric Gillespie <ep...@pretzelnet.org> wrote:
> No, we've tried at Google, as well.  It's got a long way to go.
> That's my point: as long as it sits there unused, it's not going
> to get any better.  As a matter of fact, it seems likely to me
> that it's only going to get worse.

I kinda doubt it's got a 'long way to go' - it passes all of the
regression tests on trunk - or did the last time I checked.  True,
ra_serf may need some polishing or a feature here or there - but I
don't think it's going to need *massive* amounts of work - just people
who are willing to add the features they think are missing before it
is acceptable to them.  But, it shouldn't need a tremendous amount of
effort as all of the hard stuff is already done...

FWIW, on serf-dev, Lieven said that he was looking into SSPI and that
it looked fairly straightforward to add.  *shrug*

Again, ra_serf 'works for me' - I use it without issues.  I'd advocate
flipping the default on trunk and any lingering bugs will get worked
out really fast.  =)  For those who really want the features that only
neon provides, it'll still be there ready to be used - I don't see the
harm in trying it.  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Eric Gillespie <ep...@pretzelnet.org>.
"David Glasser" <gl...@mit.edu> writes:

> On 6/8/07, Eric Gillespie <ep...@pretzelnet.org> wrote:
> > Whatever happened to making serf the default DAV layer in 1.5?
> > Last i heard, we were going to follow fsfs in this respect.  fsfs
> > was an option in 1.1 and the default in 1.2.  Serf was an option
> > in 1.4, shouldn't we make it the default now?
> >
> > If we don't switch to serf, it's never going to get the kind of
> > attention it needs...
> 
> Are many people running it as their day-to-day implementation?  I
> recently tried to run tests for a patch I was working on over serf,
> and found several places where it handled errors (like invalid user
> input or network problems, not internal corruption) with aborts... I
> fixed one or two of those places, but it definitely didn't fill me
> with confidence that ra_serf, in its current incarnation, is ready to
> be our primary implementation.

No, we've tried at Google, as well.  It's got a long way to go.
That's my point: as long as it sits there unused, it's not going
to get any better.  As a matter of fact, it seems likely to me
that it's only going to get worse.

--  
Eric Gillespie <*> epg@pretzelnet.org

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by David Glasser <gl...@mit.edu>.
On 6/8/07, Eric Gillespie <ep...@pretzelnet.org> wrote:
> Whatever happened to making serf the default DAV layer in 1.5?
> Last i heard, we were going to follow fsfs in this respect.  fsfs
> was an option in 1.1 and the default in 1.2.  Serf was an option
> in 1.4, shouldn't we make it the default now?
>
> If we don't switch to serf, it's never going to get the kind of
> attention it needs...

Are many people running it as their day-to-day implementation?  I
recently tried to run tests for a patch I was working on over serf,
and found several places where it handled errors (like invalid user
input or network problems, not internal corruption) with aborts... I
fixed one or two of those places, but it definitely didn't fill me
with confidence that ra_serf, in its current incarnation, is ready to
be our primary implementation.  (In my copious free time I'd like to
learn how serf itself works so I can be more help in fixing these
issues...)

--dave

-- 
David Glasser | glasser@mit.edu | http://www.davidglasser.net/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On 6/11/07, C. Michael Pilato <cm...@collab.net> wrote:
> That allowed me to get a build of Subversion for which serf was activated.
> I can checkout, update, and commit with this client.  But import segfaults,
> and that's keeping me from even running the test suite.  Here's a stack trace.

Well - that's because you have pool debugging turned on.  It should
pass the test suite if you disable it.  Serf can interact with the
pools in a way that APR doesn't like (as it accesses the allocators
directly).  Yes, there *might* be an underlying problem here, but it's
also just as likely to be a red herring.  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Serf as default DAV client?

Posted by "C. Michael Pilato" <cm...@collab.net>.
Eric Gillespie wrote:
> Whatever happened to making serf the default DAV layer in 1.5? Last i
> heard, we were going to follow fsfs in this respect.  fsfs was an option
> in 1.1 and the default in 1.2.  Serf was an option in 1.4, shouldn't we
> make it the default now?
> 
> If we don't switch to serf, it's never going to get the kind of attention
> it needs...

Inspired by this mail, I've tried to get serf going on over here instead of
neon.

I ran quickly into a bug in serfmake, which Justin kindly fixed.

That allowed me to get a build of Subversion for which serf was activated.
I can checkout, update, and commit with this client.  But import segfaults,
and that's keeping me from even running the test suite.  Here's a stack trace.

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread -1215985984 (LWP 23247)]
0xb7a01c7e in apr_allocator_free (allocator=0x41414141, node=0x809f898)
    at memory/unix/apr_pools.c:319
(gdb) where
#0  0xb7a01c7e in apr_allocator_free (allocator=0x41414141, node=0x809f898)
    at memory/unix/apr_pools.c:319
#1  0xb7d37c6b in serf_bucket_mem_free (allocator=0x809b588, block=0x809f8b8)
    at buckets/allocator.c:238
#2  0xb7d3660c in serf_default_destroy_and_data (bucket=0x809b7f8)
    at buckets/buckets.c:117
#3  0xb7d36e63 in serf_response_destroy_and_data (bucket=0x809b7f8)
    at buckets/response_buckets.c:96
#4  0xb7d34515 in cancel_request (request=0x80953b8, list=0x8092b64,
    notify_request=<value optimized out>) at ./context.c:368
#5  0xb7d34778 in serf_connection_close (conn=0x8092b30) at ./context.c:1019
#6  0xb7d56592 in svn_ra_serf__cleanup_serf_session (data=0x8091370)
    at subversion/libsvn_ra_serf/util.c:143
#7  0xb7a02878 in pool_clear_debug (pool=0x8091420,
    file_line=<value optimized out>) at memory/unix/apr_pools.c:2034
#8  0xb7a02b47 in pool_destroy_debug (pool=0x8091420,
    file_line=0xb7ee2cfc "subversion/libsvn_client/commit.c:690")
    at memory/unix/apr_pools.c:1457
#9  0xb7a02851 in pool_clear_debug (pool=0x8090dd8,
    file_line=0xb7ee2cfc "subversion/libsvn_client/commit.c:690")
    at memory/unix/apr_pools.c:1369
#10 0xb7a02c74 in apr_pool_clear_debug (pool=0x8090dd8,
    file_line=0xb7ee2cfc "subversion/libsvn_client/commit.c:690")
    at memory/unix/apr_pools.c:1432
#11 0xb7ebc2eb in svn_client_import2 (commit_info_p=0xbfcc7d00,
    path=0x80909f0 "",
    url=0x8090a60 "http://svn.collab.net/tests/repos/serf-import-1",
    nonrecursive=0, no_ignore=0, ctx=0x80757c0, pool=0x8074bc8)
    at subversion/libsvn_client/commit.c:690
#12 0x0804f297 in svn_cl__import (os=0x8074f50, baton=0xbfcc7f08,
    pool=0x8074bc8) at subversion/svn/import-cmd.c:115
#13 0x080549b7 in main (argc=6, argv=0xbfcc8024) at subversion/svn/main.c:1689


-- 
C. Michael Pilato <cm...@collab.net>
CollabNet   <>   www.collab.net   <>   Distributed Development On Demand