You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Stefan Sperling <st...@elego.de> on 2008/05/29 08:45:40 UTC

commits made directly to 1.5.x (was: Re: svn commit: r31506 - branches/1.5.x/subversion/svn)

On Wed, May 28, 2008 at 05:58:14PM -0400, Mark Phippard wrote:
> On Wed, May 28, 2008 at 5:22 PM, Lieven Govaerts <sv...@mobsol.be> wrote:
> > As simple as this change might be, it breaks one of the getopt tests. The getopt tests are one of the only tests really matching the actual
> > string content character per character. I know these tests are overly strict, but while we have them I propose you run at least those
> > before you commit a text change inside the code.
> 
> And as you said in the other email, why are we doing direct commits to
> the branch, without going through STATUS or even a dev@ email, right
> before we are about to make the release?

Shouldn't we immediately back out r31503 and r31506, both of which were
directly committed to 1.5.x without being voted for in STATUS?
Because what's the point in having STATUS if people don't use it?

http://subversion.tigris.org/hacking.html#release-stabilization
clearly says:

  Core code changes, of course, require voting, and restart the
  soak or test period, since otherwise the change could be undertested.

Both changes committed to 1.5.x fall into the 'core code changes'
category, one actually broke a test.

I'm not sure whether *all* changes have to go through trunk first,
but so far I haven't seen any change not going through trunk.
Please enlighten me if I'm wrong on this one.

I personally think having changes go through trunk first is very,
very valuable. Because a consistent direction of the flow of changes
we make helps everyone understand what's happening. Of course,
this excludes trivial release-specific stuff like setting version
numbers on a branch, but other than that I don't see any reason why
a change should not enter our code base through trunk. It will trickle
down into a release branch eventually if it needs to.

Stefan

Re: commits made directly to 1.5.x

Posted by Arfrever Frehtes Taifersar Arahesis <ar...@gmail.com>.
2008-05-29 20:01:20 Lieven Govaerts napisaƂ(a):
> Stefan Sperling wrote:
> > On Wed, May 28, 2008 at 05:58:14PM -0400, Mark Phippard wrote:
> >   
> >> On Wed, May 28, 2008 at 5:22 PM, Lieven Govaerts <sv...@mobsol.be> wrote:
> >>     
> >>> As simple as this change might be, it breaks one of the getopt tests. The getopt tests are one of the only tests really matching the actual
> >>> string content character per character. I know these tests are overly strict, but while we have them I propose you run at least those
> >>> before you commit a text change inside the code.
> >>>       
> >> And as you said in the other email, why are we doing direct commits to
> >> the branch, without going through STATUS or even a dev@ email, right
> >> before we are about to make the release?
> >>     
> >
> > Shouldn't we immediately back out r31503 and r31506, both of which were
> > directly committed to 1.5.x without being voted for in STATUS?
> 
> While I don't think r31503 and r31506 were urgent 1.5.x material, they
> don't do much harm either. They're included in RC8 now, so the cost of
> rolling them back (ie; RC9) will be bigger than the benefit.
> 
> Let's just avoid this situation in the future.

+1

-- 
Arfrever Frehtes Taifersar Arahesis

Re: commits made directly to 1.5.x

Posted by Stefan Sperling <st...@elego.de>.
On Fri, May 30, 2008 at 02:09:54PM +0200, Jens Seidel wrote:
> Hi Stefan,
> 
> On Fri, May 30, 2008 at 01:36:11PM +0200, Stefan Sperling wrote:
> > On Thu, May 29, 2008 at 09:18:35PM +0200, Jens Seidel wrote:
> > > The problem was that xgettext, a tool to extract messages from source
> > > code to put it into a .pot translation template file, failed to parse
> > > these strings containing macros. This resulted in no such messages
> > > in the translation PO files (and were that's why not translated) which I
> > > fixed by manually inserting these messages.
> > > Without macros xgettext works well and even after the next po-update.py
> > > call PO files no messages will be missing.
> 
> I noticed that there seems some confusion about the patches, that's why I
> explained it in a little bit more detail.
> 
> > Including this kind of information in your log messages would
> > be a very, very nice thing to do. In the log, you only mention "xgettext
> > problem" and "xgettext limitation". That provides no information at all
> 
> The log for r31484 contains:
> Use workaround for an xgettext limitation.
> 
> * subversion/svn/main.c,
>   tools/po/po-update.sh,
>   configure.ac:
>   xgettext fails to look up macros in some messages so that these are not put
>   into PO files and that's why untranslated. This patch explicitly adds
>   duplicates of these strings with manually expanded macros.

Ah, there it is. Sorry, I was looking at the 1.5.x logs, only.
The log message on trunk is fine.

The 1.5.x ones I was looking at are:

------------------------------------------------------------------------
r31495 | jensseidel | 2008-05-28 15:06:36 +0200 (Wed, 28 May 2008) | 3
lines

Added another missing message into PO files. It occurs in the "resolve" help
and was again missing because of an xgettext limitation.



------------------------------------------------------------------------
r31450 | jensseidel | 2008-05-26 16:44:31 +0200 (Mon, 26 May 2008) | 6 lines

German translation update.

* subversion/po/de.po: Current status: 1804t 0f 0u
  Three untranslated messages were added, two were missing because of a
  xgettext problem. Most text fits now on a 80 character wide terminal.


Stefan

Re: commits made directly to 1.5.x

Posted by Jens Seidel <je...@users.sourceforge.net>.
Hi Stefan,

On Fri, May 30, 2008 at 01:36:11PM +0200, Stefan Sperling wrote:
> On Thu, May 29, 2008 at 09:18:35PM +0200, Jens Seidel wrote:
> > The problem was that xgettext, a tool to extract messages from source
> > code to put it into a .pot translation template file, failed to parse
> > these strings containing macros. This resulted in no such messages
> > in the translation PO files (and were that's why not translated) which I
> > fixed by manually inserting these messages.
> > Without macros xgettext works well and even after the next po-update.py
> > call PO files no messages will be missing.

I noticed that there seems some confusion about the patches, that's why I
explained it in a little bit more detail.

> Including this kind of information in your log messages would
> be a very, very nice thing to do. In the log, you only mention "xgettext
> problem" and "xgettext limitation". That provides no information at all

The log for r31484 contains:
Use workaround for an xgettext limitation.

* subversion/svn/main.c,
  tools/po/po-update.sh,
  configure.ac:
  xgettext fails to look up macros in some messages so that these are not put
  into PO files and that's why untranslated. This patch explicitly adds
  duplicates of these strings with manually expanded macros.

r31496 is shorter:

Marked a further string explicitly translatable which is also missed by
xgettext.

* subversion/svn/main.c:
  Added additional GETTEXT_NOOP macro. Also use consistent comments near
  other affected strings.

Maybe I should have added a "Follow-up to r31496" is it. Sorry for
omitting it. Nevertheless "further" indicates that there was a similar
commit.

> to the reader. Not everyone knows what xgettext is, let alone what

Mhm, OK. But also not everyone knows apr and it isn't always explained :-)
I assumed indeed that it is known (at least for the initial committer).

Just asking for details after reading
http://subversion.tigris.org/servlets/ReadMsg?list=dev&msgNo=139147
had maybe avoid some confusion too ...

By the way, the direct reply of this mail confused me as I missed apr
knowledge.

> its quirks are. Could you propedit the log messages for the revisions
> where you fixed the problem?

I could and incorporated already previous comments from you (by adding
Approved-by: tags, ...) but here I'm not sure how to proceed. Please
change it yourself.

> And, by the way, what you wrote above is also a good example for
> what you could have sent to the list in order to discuss the fix before
> committing.

I did so with a previous patch. Since I also planned at the beginning
to apply it later to 1.5.x as well (after getting permission!) to fix
the three untranslated messages and since there was nearly no further
time to wait for comments because of the release I tried to risk
committing it to trunk. Please note that I honoured the (bogus?) comment
I got to my mail by changing N_ into GETTEXT_NOOP.


Jens

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

Re: commits made directly to 1.5.x

Posted by Stefan Sperling <st...@elego.de>.
On Thu, May 29, 2008 at 09:18:35PM +0200, Jens Seidel wrote:
> The problem was that xgettext, a tool to extract messages from source
> code to put it into a .pot translation template file, failed to parse
> these strings containing macros. This resulted in no such messages
> in the translation PO files (and were that's why not translated) which I
> fixed by manually inserting these messages.
> Without macros xgettext works well and even after the next po-update.py
> call PO files no messages will be missing.

Including this kind of information in your log messages would
be a very, very nice thing to do. In the log, you only mention "xgettext
problem" and "xgettext limitation". That provides no information at all
to the reader. Not everyone knows what xgettext is, let alone what
its quirks are. Could you propedit the log messages for the revisions
where you fixed the problem?

And, by the way, what you wrote above is also a good example for
what you could have sent to the list in order to discuss the fix before
committing. I know I've already raised this issue and we talked it
through in another thread, but the above is a really good explanation
of *why* you made the change, so I'm pointing it out again here. The above
is exactly the kind of information others need to understand and evaluate
your changes.

> > > Let's just avoid this situation in the future.
> 
> Right! Don't revert it! It was carefully proven by me and others!

OK.

I agree with Lieven, -rc8 is out of the bag, so we can just
leave them in now.

Stefan

Re: commits made directly to 1.5.x

Posted by Daniel Shahaf <d....@daniel.shahaf.co.il>.
Jens Seidel wrote on Thu, 29 May 2008 at 21:18 +0200:
> On Thu, May 29, 2008 at 09:16:39PM +0300, Daniel Shahaf wrote:
> > Lieven Govaerts wrote on Thu, 29 May 2008 at 20:01 +0200:
> > > Stefan Sperling wrote:
> Great! If you would revert r31506 you would add a regression because of
> r31501.
> 
> > > While I don't think r31503 and r31506 were urgent 1.5.x material, they
> > > don't do much harm either. They're included in RC8 now, so the cost of
> > > rolling them back (ie; RC9) will be bigger than the benefit.
> > > 
> > 
> > We can include them in 1.5.1, as they only change the 'svn --help'
> > output.  No rc9.
> 
> That's wrong again! These patches just manually expand macros in help
> outputs (for commands resolve, mergeinfo and switch). There should be
> no change in the output!
> 

That's what I said.

> The problem was that xgettext, a tool to extract messages from source
> code to put it into a .pot translation template file, failed to parse
> these strings containing macros. This resulted in no such messages
> in the translation PO files (and were that's why not translated) which I
> fixed by manually inserting these messages.
> Without macros xgettext works well and even after the next po-update.py
> call PO files no messages will be missing.
> 
> > > Let's just avoid this situation in the future.
> 
> Right! Don't revert it! It was carefully proven by me and others!
> 
> Jens
> 

Too many exclamation marks.



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

Re: commits made directly to 1.5.x

Posted by Jens Seidel <je...@users.sourceforge.net>.
On Thu, May 29, 2008 at 09:16:39PM +0300, Daniel Shahaf wrote:
> Lieven Govaerts wrote on Thu, 29 May 2008 at 20:01 +0200:
> > Stefan Sperling wrote:
> > > On Wed, May 28, 2008 at 05:58:14PM -0400, Mark Phippard wrote:
> > >   
> > >> On Wed, May 28, 2008 at 5:22 PM, Lieven Govaerts <sv...@mobsol.be> wrote:
> > >>     
> > >>> As simple as this change might be, it breaks one of the getopt tests. The getopt tests are one of the only tests really matching the actual
> > >>> string content character per character. I know these tests are overly strict, but while we have them I propose you run at least those
> > >>> before you commit a text change inside the code.
> > >>>       
> > >> And as you said in the other email, why are we doing direct commits to
> > >> the branch, without going through STATUS or even a dev@ email, right
> > >> before we are about to make the release?
> > >>     
> > >
> > > Shouldn't we immediately back out r31503 and r31506, both of which were
> > > directly committed to 1.5.x without being voted for in STATUS?

Great! If you would revert r31506 you would add a regression because of
r31501.

> > While I don't think r31503 and r31506 were urgent 1.5.x material, they
> > don't do much harm either. They're included in RC8 now, so the cost of
> > rolling them back (ie; RC9) will be bigger than the benefit.
> > 
> 
> We can include them in 1.5.1, as they only change the 'svn --help'
> output.  No rc9.

That's wrong again! These patches just manually expand macros in help
outputs (for commands resolve, mergeinfo and switch). There should be
no change in the output!

The problem was that xgettext, a tool to extract messages from source
code to put it into a .pot translation template file, failed to parse
these strings containing macros. This resulted in no such messages
in the translation PO files (and were that's why not translated) which I
fixed by manually inserting these messages.
Without macros xgettext works well and even after the next po-update.py
call PO files no messages will be missing.

> > Let's just avoid this situation in the future.

Right! Don't revert it! It was carefully proven by me and others!

Jens

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

Re: commits made directly to 1.5.x

Posted by Daniel Shahaf <d....@daniel.shahaf.co.il>.
Lieven Govaerts wrote on Thu, 29 May 2008 at 20:01 +0200:
> Stefan Sperling wrote:
> > On Wed, May 28, 2008 at 05:58:14PM -0400, Mark Phippard wrote:
> >   
> >> On Wed, May 28, 2008 at 5:22 PM, Lieven Govaerts <sv...@mobsol.be> wrote:
> >>     
> >>> As simple as this change might be, it breaks one of the getopt tests. The getopt tests are one of the only tests really matching the actual
> >>> string content character per character. I know these tests are overly strict, but while we have them I propose you run at least those
> >>> before you commit a text change inside the code.
> >>>       
> >> And as you said in the other email, why are we doing direct commits to
> >> the branch, without going through STATUS or even a dev@ email, right
> >> before we are about to make the release?
> >>     
> >
> > Shouldn't we immediately back out r31503 and r31506, both of which were
> > directly committed to 1.5.x without being voted for in STATUS?
> 
> While I don't think r31503 and r31506 were urgent 1.5.x material, they
> don't do much harm either. They're included in RC8 now, so the cost of
> rolling them back (ie; RC9) will be bigger than the benefit.
> 

We can include them in 1.5.1, as they only change the 'svn --help'
output.  No rc9.

> Let's just avoid this situation in the future.
> 
> Lieven
> 

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

Re: commits made directly to 1.5.x

Posted by Lieven Govaerts <sv...@mobsol.be>.
Stefan Sperling wrote:
> On Wed, May 28, 2008 at 05:58:14PM -0400, Mark Phippard wrote:
>   
>> On Wed, May 28, 2008 at 5:22 PM, Lieven Govaerts <sv...@mobsol.be> wrote:
>>     
>>> As simple as this change might be, it breaks one of the getopt tests. The getopt tests are one of the only tests really matching the actual
>>> string content character per character. I know these tests are overly strict, but while we have them I propose you run at least those
>>> before you commit a text change inside the code.
>>>       
>> And as you said in the other email, why are we doing direct commits to
>> the branch, without going through STATUS or even a dev@ email, right
>> before we are about to make the release?
>>     
>
> Shouldn't we immediately back out r31503 and r31506, both of which were
> directly committed to 1.5.x without being voted for in STATUS?

While I don't think r31503 and r31506 were urgent 1.5.x material, they
don't do much harm either. They're included in RC8 now, so the cost of
rolling them back (ie; RC9) will be bigger than the benefit.

Let's just avoid this situation in the future.

Lieven


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

Re: commits made directly to 1.5.x (was: Re: svn commit: r31506 - branches/1.5.x/subversion/svn)

Posted by Daniel Shahaf <d....@daniel.shahaf.co.il>.
Erik Huelsmann wrote on Thu, 29 May 2008 at 11:06 +0200:
> On 5/29/08, Stefan Sperling <st...@elego.de> wrote:
> > On Wed, May 28, 2008 at 05:58:14PM -0400, Mark Phippard wrote:
> > > On Wed, May 28, 2008 at 5:22 PM, Lieven Govaerts <sv...@mobsol.be> wrote:
> > > > As simple as this change might be, it breaks one of the getopt tests. The getopt tests are one of the only tests really matching the actual
> > > > string content character per character. I know these tests are overly strict, but while we have them I propose you run at least those
> > > > before you commit a text change inside the code.
> > >
> > > And as you said in the other email, why are we doing direct commits to
> > > the branch, without going through STATUS or even a dev@ email, right
> > > before we are about to make the release?
> >
> > Shouldn't we immediately back out r31503 and r31506, both of which were
> > directly committed to 1.5.x without being voted for in STATUS?
> > Because what's the point in having STATUS if people don't use it?
> >

+1.

It didn't go through STATUS and I saw no discussion on the list either.


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

Re: commits made directly to 1.5.x (was: Re: svn commit: r31506 - branches/1.5.x/subversion/svn)

Posted by Erik Huelsmann <eh...@gmail.com>.
On 5/29/08, Stefan Sperling <st...@elego.de> wrote:
> On Wed, May 28, 2008 at 05:58:14PM -0400, Mark Phippard wrote:
> > On Wed, May 28, 2008 at 5:22 PM, Lieven Govaerts <sv...@mobsol.be> wrote:
> > > As simple as this change might be, it breaks one of the getopt tests. The getopt tests are one of the only tests really matching the actual
> > > string content character per character. I know these tests are overly strict, but while we have them I propose you run at least those
> > > before you commit a text change inside the code.
> >
> > And as you said in the other email, why are we doing direct commits to
> > the branch, without going through STATUS or even a dev@ email, right
> > before we are about to make the release?
>
> Shouldn't we immediately back out r31503 and r31506, both of which were
> directly committed to 1.5.x without being voted for in STATUS?
> Because what's the point in having STATUS if people don't use it?
>
> http://subversion.tigris.org/hacking.html#release-stabilization
> clearly says:
>
>  Core code changes, of course, require voting, and restart the
>  soak or test period, since otherwise the change could be undertested.
>
> Both changes committed to 1.5.x fall into the 'core code changes'
> category, one actually broke a test.
>
> I'm not sure whether *all* changes have to go through trunk first,
> but so far I haven't seen any change not going through trunk.
> Please enlighten me if I'm wrong on this one.

You're not wrong. Going through trunk first makes sure all changes end
up in all versions beyond 1.5. The only time a change can be committed
to a branch (but never directly) is when the code has been removed or
rewritten on trunk.

> I personally think having changes go through trunk first is very,
> very valuable. Because a consistent direction of the flow of changes
> we make helps everyone understand what's happening. Of course,
> this excludes trivial release-specific stuff like setting version
> numbers on a branch, but other than that I don't see any reason why
> a change should not enter our code base through trunk. It will trickle
> down into a release branch eventually if it needs to.

Exactly. A clear flow of changes also guarantees future quality of the
software as described above.

Bye,

Erik.

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