You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by David Glasser <gl...@davidglasser.net> on 2008/06/08 06:11:06 UTC
you're going to kill me, but... [Re: svn commit: r31625 - branches/1.5.x]
Um, I really don't want to drag this process out any longer. But as
far as I can tell, without backporting these revisions, every single
program written against the svn_repos API before 1.5.x (ie, the API
which does not include svn_repos_set_repos_capabilities) which tries
to commit anything to a repository with a start-commit hook will
dereference NULL (ie, crash) every time.
Please please please somebody convince me that I am wrong in that
conclusion above. Because if we actually claim to care about API
users, that's pretty serious.
--dave
On Sat, Jun 7, 2008 at 11:08 PM, <gl...@tigris.org> wrote:
> Author: glasser
> Date: Sat Jun 7 23:08:11 2008
> New Revision: 31625
>
> Log:
> Vote for r31620/31622, and note that it may be important.
>
> Modified:
> branches/1.5.x/STATUS
>
> Modified: branches/1.5.x/STATUS
> URL: http://svn.collab.net/viewvc/svn/branches/1.5.x/STATUS?pathrev=31625&r1=31624&r2=31625
> ==============================================================================
> --- branches/1.5.x/STATUS Sat Jun 7 11:15:11 2008 (r31624)
> +++ branches/1.5.x/STATUS Sat Jun 7 23:08:11 2008 (r31625)
> @@ -110,10 +110,15 @@ Candidate changes for 1.5.1:
> * r31620, r31622
> Don't blindly assume that the client will always report capabilities.
> Fixes a segfault in svn_repos__hooks_start_commit(). svn(1) won't
> - trigger it, but other clients could. This can wait for 1.5.1 on the
> - same grounds as r31223.
> + trigger it, but other clients could.
> + Notes:
> + stsp: This can wait for 1.5.1 on the same grounds as r31223.
> + glasser: Um, doesn't not applying this segfault *every single
> + program* written against the pre-1.5 svn_repos API if
> + the repository has a start-commit hook? As much as I
> + hate to drag out this process...
> Votes:
> - +1: stsp, danielsh
> + +1: stsp, danielsh, glasser
>
> Approved changes (all releases):
> ================================
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: svn-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: svn-help@subversion.tigris.org
>
>
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: you're going to kill me, but... [Re: svn commit: r31625 - branches/1.5.x]
Posted by David Glasser <gl...@davidglasser.net>.
On Sun, Jun 8, 2008 at 10:10 PM, Karl Fogel <kf...@red-bean.com> wrote:
> "Mark Phippard" <ma...@gmail.com> writes:
>> David Glasser <gl...@davidglasser.net> wrote:
>>> Um, I really don't want to drag this process out any longer. But as
>>> far as I can tell, without backporting these revisions, every single
>>> program written against the svn_repos API before 1.5.x (ie, the API
>>> which does not include svn_repos_set_repos_capabilities) which tries
>>> to commit anything to a repository with a start-commit hook will
>>> dereference NULL (ie, crash) every time.
>>>
>>> Please please please somebody convince me that I am wrong in that
>>> conclusion above. Because if we actually claim to care about API
>>> users, that's pretty serious.
>>
>> [...]
>>
>> I have had several fellow committers take indirect swipes at me on IRC
>> or in this list. Saying that I do not understand what a release
>> candidate means etc. It is pretty clear to me though, that if we were
>> not going through this process and "threatening" to release then the
>> software would never get tested. The last few bugs reported make that
>> clear. This bug in particular has been sitting in the code since
>> November 2007. I am assuming this bug impacts SVK, since clkao
>> reported it and I have seen some other talk of segfaults from SVK
>> users. Does this mean that no one has tried SVK with 1.5 code until
>> the last week? If that is true, are we saying they cannot wait until
>> 1.5.1? I've been testing the Java API daily since last summer.
>
> I think David's last sentence was just a strongly-worded way of saying
> what he thinks our best course of action is, nothing more -- no swipes
> at anyone. (At least, that's how it looked from here!)
Indeed, no personal swipes meant. And I was serious... I really just
wanted to be told "nah, this doesn't actually break as much as it
looks like it does, let's release".
--dave
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: you're going to kill me, but... [Re: svn commit: r31625 - branches/1.5.x]
Posted by Mark Phippard <ma...@gmail.com>.
On Mon, Jun 9, 2008 at 1:10 AM, Karl Fogel <kf...@red-bean.com> wrote:
>> I have had several fellow committers take indirect swipes at me on IRC
>> or in this list. Saying that I do not understand what a release
>> candidate means etc. It is pretty clear to me though, that if we were
>> not going through this process and "threatening" to release then the
>> software would never get tested. The last few bugs reported make that
>> clear. This bug in particular has been sitting in the code since
>> November 2007. I am assuming this bug impacts SVK, since clkao
>> reported it and I have seen some other talk of segfaults from SVK
>> users. Does this mean that no one has tried SVK with 1.5 code until
>> the last week? If that is true, are we saying they cannot wait until
>> 1.5.1? I've been testing the Java API daily since last summer.
>
> I think David's last sentence was just a strongly-worded way of saying
> what he thinks our best course of action is, nothing more -- no swipes
> at anyone. (At least, that's how it looked from here!)
I am sorry too. That was just a poorly worded segue to a topic I
wanted to bring up at some point. I did not mean to imply David was
doing that, or even had done that in the past. As I said, I also do
not disagree with the need to backport this. I do disagree with any
notion that we are compelled to fix every bug we find before we can
ship. Clearly that is not the case and there are plenty of bugs more
significant than some of these that we have not even tried to fix.
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: you're going to kill me, but... [Re: svn commit: r31625 - branches/1.5.x]
Posted by Karl Fogel <kf...@red-bean.com>.
"Mark Phippard" <ma...@gmail.com> writes:
> David Glasser <gl...@davidglasser.net> wrote:
>> Um, I really don't want to drag this process out any longer. But as
>> far as I can tell, without backporting these revisions, every single
>> program written against the svn_repos API before 1.5.x (ie, the API
>> which does not include svn_repos_set_repos_capabilities) which tries
>> to commit anything to a repository with a start-commit hook will
>> dereference NULL (ie, crash) every time.
>>
>> Please please please somebody convince me that I am wrong in that
>> conclusion above. Because if we actually claim to care about API
>> users, that's pretty serious.
>
> [...]
>
> I have had several fellow committers take indirect swipes at me on IRC
> or in this list. Saying that I do not understand what a release
> candidate means etc. It is pretty clear to me though, that if we were
> not going through this process and "threatening" to release then the
> software would never get tested. The last few bugs reported make that
> clear. This bug in particular has been sitting in the code since
> November 2007. I am assuming this bug impacts SVK, since clkao
> reported it and I have seen some other talk of segfaults from SVK
> users. Does this mean that no one has tried SVK with 1.5 code until
> the last week? If that is true, are we saying they cannot wait until
> 1.5.1? I've been testing the Java API daily since last summer.
I think David's last sentence was just a strongly-worded way of saying
what he thinks our best course of action is, nothing more -- no swipes
at anyone. (At least, that's how it looked from here!)
By the way, the original bug is mine, sorry about that.
> I do not want to sidetrack the discussion and further delay the
> release. We might as well roll rc10 once the fix is backported (and
> possibly the other one Stefan Sperling mentioned) and we can start
> talking about 1.5.0-final again in a week when Hyrum returns from
> whatever he is doing.
Fortunately, Hyrum's absence would coincide with the soak time needed by
that final RC anyway.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: you're going to kill me, but... [Re: svn commit: r31625 - branches/1.5.x]
Posted by Chia-liang Kao <cl...@clkao.org>.
2008/6/8 Mark Phippard <ma...@gmail.com>:
> November 2007. I am assuming this bug impacts SVK, since clkao
> reported it and I have seen some other talk of segfaults from SVK
> users. Does this mean that no one has tried SVK with 1.5 code until
> the last week? If that is true, are we saying they cannot wait until
> 1.5.1? I've been testing the Java API daily since last summer.
FTR, this does not cause SVK regression unless you have a start-commit
hook, which usually isn't the case. And the problem was found when
testing the SVN::Hook module which isn't really widely used. One of
its regression test happens to use the start-commit hook and caused
segfault.
Re: you're going to kill me, but... [Re: svn commit: r31625 - branches/1.5.x]
Posted by Mark Phippard <ma...@gmail.com>.
On Sun, Jun 8, 2008 at 2:11 AM, David Glasser <gl...@davidglasser.net> wrote:
> Um, I really don't want to drag this process out any longer. But as
> far as I can tell, without backporting these revisions, every single
> program written against the svn_repos API before 1.5.x (ie, the API
> which does not include svn_repos_set_repos_capabilities) which tries
> to commit anything to a repository with a start-commit hook will
> dereference NULL (ie, crash) every time.
>
> Please please please somebody convince me that I am wrong in that
> conclusion above. Because if we actually claim to care about API
> users, that's pretty serious.
That seems like a "loaded" way to phrase the problem. If we didn't
care about API users we wouldn't fix the bugs in the first place, or
be willing to backport them. Obviously we care. We have several bugs
that have been in the product since 1.0 that make Subversion all but
impossible to use for certain classes of OSX users. Does that mean we
do not care about OSX users? Actually, we have been accused of that
before, but I do not think it is true. All software has bugs but if
you never release you might as well not exist.
I have had several fellow committers take indirect swipes at me on IRC
or in this list. Saying that I do not understand what a release
candidate means etc. It is pretty clear to me though, that if we were
not going through this process and "threatening" to release then the
software would never get tested. The last few bugs reported make that
clear. This bug in particular has been sitting in the code since
November 2007. I am assuming this bug impacts SVK, since clkao
reported it and I have seen some other talk of segfaults from SVK
users. Does this mean that no one has tried SVK with 1.5 code until
the last week? If that is true, are we saying they cannot wait until
1.5.1? I've been testing the Java API daily since last summer.
Anyway, I am not against rolling another RC. I understand the
motivation for wanting to do so. I just do not agree with the general
notion that just because we find a bug in the API we have to pull the
emergency brake or it means we do not care. We have no problem making
the 99+% of our users that will not be impacted by any of these bugs
continue to wait for 1.5 (which includes a hell of a lot of bug fixes
by the way). But the very small number of users that are impacted by
these bugs cannot wait for a 1.5.1 (assuming they cannot apply the
patch or move to the new 1.5 API)?
I do not want to sidetrack the discussion and further delay the
release. We might as well roll rc10 once the fix is backported (and
possibly the other one Stefan Sperling mentioned) and we can start
talking about 1.5.0-final again in a week when Hyrum returns from
whatever he is doing.
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org