You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by "D.J. Heap" <dj...@gmail.com> on 2005/12/22 02:23:52 UTC
Annotate broken on trunk on Win32
Annotate/blame seems to be broken on trunk on Windows -- the 1.3
branch works fine.
Here's what I get:
C:\Projects\Subversion>svn annotate
http://svn.collab.net/repos/svn/trunk/build.conf
svn: File 'C:\Temp\tmp\tmp.2.tmp' has inconsistent newlines
svn: Inconsistent line ending style
I'll investigate deeper in a few days when I can.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "Peter N. Lundblad" <pe...@famlundblad.se>.
On Fri, 10 Feb 2006, D.J. Heap wrote:
> On 2/10/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
> [snip]
> > r17624 was what I was thinking of. Maybe that didn't do what I thought
> > it did.
>
> Ah, yes...hmm, if that patch was supposed to make diff ignore eol's
> then it doesn't seem to be working right. It looks like maybe it was
> intended to handle CRLF combos that cross buffers, but I'm not
> positive about that.
>
Just FYI, that commit maed the diff algorithm support CR as a line
terminator. Not that many people seem to use that (Mac OS 9 folks, I've
been told), but since we say we support CR line endings...
//Peter
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/10/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
[snip]
> r17624 was what I was thinking of. Maybe that didn't do what I thought
> it did.
Ah, yes...hmm, if that patch was supposed to make diff ignore eol's
then it doesn't seem to be working right. It looks like maybe it was
intended to handle CRLF combos that cross buffers, but I'm not
positive about that.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Fri, Feb 10, 2006 at 08:49:45AM -0700, D.J. Heap wrote:
> On 2/10/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
> >
> > I thought lundblad recently made changes to diff to make it
> > eol-insensitive? (i.e. diffing the two revisions in your example would
> > show no differences). Blame uses the same code, I thought, so should
> > do the same, shouldn't it?
> >
>
>
> I think it's still experimental and in discussion in the "[PATCH]
> Experimental svn diff -w" thread, and I think it's only an option,
> isn't it?
>
r17624 was what I was thinking of. Maybe that didn't do what I thought
it did.
Regards,
Malcolm
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/10/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
>
> I thought lundblad recently made changes to diff to make it
> eol-insensitive? (i.e. diffing the two revisions in your example would
> show no differences). Blame uses the same code, I thought, so should
> do the same, shouldn't it?
>
I think it's still experimental and in discussion in the "[PATCH]
Experimental svn diff -w" thread, and I think it's only an option,
isn't it?
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Fri, Feb 10, 2006 at 08:34:18AM -0700, D.J. Heap wrote:
> Hopefully my other message explained the issue better -- if we
> consider it a bug, then I think we'd have to decide how blame should
> behave when a eol-style change is made. Is that considered
> all-content-changed or no-content-changed? Right now it's either one
> depending on the eol-style and platform.
>
I thought lundblad recently made changes to diff to make it
eol-insensitive? (i.e. diffing the two revisions in your example would
show no differences). Blame uses the same code, I thought, so should
do the same, shouldn't it?
Regards,
Malcolm
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/9/06, Julian Foad <ju...@btopenworld.com> wrote:
[snip]
>
> Huh? Could you expand on that, please. If "blame" is meant to give the same
> results on different systems, and it doesn't, that's a bug.
>
> - Julian
Hopefully my other message explained the issue better -- if we
consider it a bug, then I think we'd have to decide how blame should
behave when a eol-style change is made. Is that considered
all-content-changed or no-content-changed? Right now it's either one
depending on the eol-style and platform.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Julian Foad <ju...@btopenworld.com>.
D.J. Heap wrote:
>
> Great, thanks very much! It works as expected now.
>
> Some of the files in Subversion's repository (build.conf, for example)
> show slightly different blame output on Win32 vs Linux, but I think
> that's to be expected given the history of the client and server code
> that has built that particular repository.
Huh? Could you expand on that, please. If "blame" is meant to give the same
results on different systems, and it doesn't, that's a bug.
- Julian
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/11/06, Peter N. Lundblad <pe...@famlundblad.se> wrote:
[snip]
> I'm currently implementing an ignore_eol_style option to the internal diff
> library, which might solve this.
>
Thank you, that will be very useful!
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "Peter N. Lundblad" <pe...@famlundblad.se>.
On Fri, 10 Feb 2006, D.J. Heap wrote:
> On 2/10/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
> [snip]
> If a file is added with no eol-style and then sometime later has an
> eol-style set to 'native', then it may show different blame output on
> different platforms. The original file had the newlines of it's
> original platform and so blame works 'normally' -- but on a different
> platform the revision *after* the eol-style change will get credit for
> all the lines.
>
...
> Kind of odd, but reasonable, I guess, given how eol-style handling
> works. And with the new 'ignore eol' option it could be easily worked
> around if needed.
>
I'm currently implementing an ignore_eol_style option to the internal diff
library, which might solve this.
//Peter
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/10/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
[snip]
> > Some of the files in Subversion's repository (build.conf, for example)
> > show slightly different blame output on Win32 vs Linux, but I think
> > that's to be expected given the history of the client and server code
> > that has built that particular repository. It is working perfectly on
> > the test repositories I've been using.
> >
>
> Huh. Could you show us what you mean? Is it just that the translation
> code is hiding inconsistent newlines (and so we're seeing less changes
> on win32), or is it something more serious?
Yes, that is what I was thinking -- inconsistent newlines due to bugs
in the eol handling or mixed newlines being commited before the
eol-style was set or something. But after playing with it more, I
don't think that is happending.
If a file is added with no eol-style and then sometime later has an
eol-style set to 'native', then it may show different blame output on
different platforms. The original file had the newlines of it's
original platform and so blame works 'normally' -- but on a different
platform the revision *after* the eol-style change will get credit for
all the lines.
For example, on Win32 a blame of build.conf shows the first 3 lines
credited to '4387 striker' but on Linux they are '1 svn'. The
eol-style was changed to 'native' in r4371 which was the only change
to build.conf until r4387.
But if I add a file on Linux with no eol-style, make a few changes,
then later set the eol-style to 'CRLF', the diff will show every line
changed in that commit (as well as the eol-style change) and so blame
credits that revision with every line.
Kind of odd, but reasonable, I guess, given how eol-style handling
works. And with the new 'ignore eol' option it could be easily worked
around if needed.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Thu, Feb 09, 2006 at 05:20:46PM -0700, D.J. Heap wrote:
> On 2/9/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
> > Fix blame for files whose translated eol-style differs from the
> > untranslated style. For example, files with eol-style of 'native'
> > on Win32.
>
> Great, thanks very much! It works as expected now.
>
Ok, committed as r18412. Thanks for helping me test it!
> Some of the files in Subversion's repository (build.conf, for example)
> show slightly different blame output on Win32 vs Linux, but I think
> that's to be expected given the history of the client and server code
> that has built that particular repository. It is working perfectly on
> the test repositories I've been using.
>
Huh. Could you show us what you mean? Is it just that the translation
code is hiding inconsistent newlines (and so we're seeing less changes
on win32), or is it something more serious?
Regards,
Malcolm
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/9/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
[snip]
>
> You got it exactly right: we were still passing the untranslated 'old'
> file to add_file_blame(), so every line was coming up as changed.
> How about the attached patch?
>
> [[[
> Fix blame for files whose translated eol-style differs from the
> untranslated style. For example, files with eol-style of 'native'
> on Win32.
Great, thanks very much! It works as expected now.
Some of the files in Subversion's repository (build.conf, for example)
show slightly different blame output on Win32 vs Linux, but I think
that's to be expected given the history of the client and server code
that has built that particular repository. It is working perfectly on
the test repositories I've been using.
Thanks!
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Mon, Feb 06, 2006 at 06:18:49PM -0700, D.J. Heap wrote:
> Thanks, that did help! Blame now finishes and the lines are correct.
> However, now in the blame info (rev, author) the last person to change
> the file gets credit for every single line...it's like it line-diffing
> translated and untranslated files now?
>
You got it exactly right: we were still passing the untranslated 'old'
file to add_file_blame(), so every line was coming up as changed.
How about the attached patch?
[[[
Fix blame for files whose translated eol-style differs from the
untranslated style. For example, files with eol-style of 'native'
on Win32.
Followup to r17513.
Found by: djh
brane
* subversion/libsvn_client/blame.c
(struct file_rev_baton): Add 'last_filename_translated' to hold the
name of the translated version of the prior revision.
(window_handler): Call add_file_blame() with the name of the
translated version of the prior revision, rather than the untranslated
version. Set frb->last_filename to the untranslated version of
the current revision, rather than the translated version (important
because this will be the delta source for the next revision), and
instead set frb->last_filename_translated to the translated version.
(svn_client_blame2): Initialise frb.last_filename_translated to NULL.
]]]
Regards,
Malcolm
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/8/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
> On Mon, Feb 06, 2006 at 06:18:49PM -0700, D.J. Heap wrote:
> > Thanks, that did help! Blame now finishes and the lines are correct.
> > However, now in the blame info (rev, author) the last person to change
> > the file gets credit for every single line...it's like it line-diffing
> > translated and untranslated files now?
> >
>
> Yes, I think you're right, or perhaps a translated and double-translated file. Okay, I can actually reproduce what you're getting with my patch on Linux by forcing APR_EOL_STR to "\r\n". I wasn't sure if I'd broken something before, but now I know you're also getting that, I'll see if I can take a look.
>
Thanks! I'll have time to dig further this weekend, also, if you don't.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Mon, Feb 06, 2006 at 06:18:49PM -0700, D.J. Heap wrote:
> Thanks, that did help! Blame now finishes and the lines are correct.
> However, now in the blame info (rev, author) the last person to change
> the file gets credit for every single line...it's like it line-diffing
> translated and untranslated files now?
>
Yes, I think you're right, or perhaps a translated and double-translated file. Okay, I can actually reproduce what you're getting with my patch on Linux by forcing APR_EOL_STR to "\r\n". I wasn't sure if I'd broken something before, but now I know you're also getting that, I'll see if I can take a look.
Regards,
Malcolm
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/6/06, Malcolm Rowe <ma...@farside.org.uk> wrote:
[snip]
> Spot on. I think this was caused by the fix to issue #2431 in r17513.
> The change to libsvn_client/blame.c:window_handler() in that revision
> sets (struct delta_baton *)->filename to the translated file's name,
> just before we assign it to (struct delta_baton *)->last_filename,
> the source for the next delta.
>
> I think the attached patch should resolve that, though I can't reproduce
> the problem on Darwin - even with a file that does need translation.
>
> Having said that, it looks fairly straightforward. Can someone try this
> out on Windows and confirm whether it fixes the problem?
Thanks, that did help! Blame now finishes and the lines are correct.
However, now in the blame info (rev, author) the last person to change
the file gets credit for every single line...it's like it line-diffing
translated and untranslated files now?
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Malcolm Rowe <ma...@farside.org.uk>.
On Sat, Feb 04, 2006 at 10:21:26PM +0100, Branko Čibej wrote:
> It seems that blame is applying deltas to the translated buffer, not the
> untranslated one. That's seriously broken.
>
Spot on. I think this was caused by the fix to issue #2431 in r17513.
The change to libsvn_client/blame.c:window_handler() in that revision
sets (struct delta_baton *)->filename to the translated file's name,
just before we assign it to (struct delta_baton *)->last_filename,
the source for the next delta.
I think the attached patch should resolve that, though I can't reproduce
the problem on Darwin - even with a file that does need translation.
Having said that, it looks fairly straightforward. Can someone try this
out on Windows and confirm whether it fixes the problem?
[[[
Fix blame for translated files. Followup to r17513.
Found by: djh
brane
* subversion/libsvn_client/blame.c
(window_handler): Introduce a local variable to hold the name of
the temporary translated file, so that the delta source for the next
revision will use the untranslated file.
]]]
Incidentally, the fix for issue #2431 doesn't appear to have changed
old_blame() at all. Presumably this means the original issue is still
outstanding for 1.0.x servers?
Regards,
Malcolm
Re: Annotate broken on trunk on Win32
Posted by Branko Čibej <br...@xbc.nu>.
D.J. Heap wrote:
> On 2/4/06, D.J. Heap <dj...@gmail.com> wrote:
> [snip]
>
>> Even worse, on Windows doing:
>>
>> svn blame -r1:4657 build.conf
>>
>> gives no errors, but outputs corrupt lines and not even all the lines.
>> Blaming r1:4656 at least gives consistent output and the whole
>> file...it looks like r4657 (a one line change to build.conf) causes it
>> to flip out. But it doesn't actually throw an error until blaming
>> r1:4805 -- after that blame never works for build.conf on Windows (or
>> probably any platform where native eol is not LF), it just throws the
>> 'inconsistent eols' error.
>>
>
>
> I've narrowed this down to svn_txdelta_apply_instructions corrupting
> the target buffer -- I'm not sure exactly what that means
> though...does it mean we have bad deltas in the repository (it works
> on Linux so that seems unlikely) or that blame is incorrectly applying
> them due to eol size changes or what?
>
> Can someone with more expertise here take a look?
>
It seems that blame is applying deltas to the translated buffer, not the
untranslated one. That's seriously broken.
-- Brane
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/4/06, D.J. Heap <dj...@gmail.com> wrote:
[snip]
> Even worse, on Windows doing:
>
> svn blame -r1:4657 build.conf
>
> gives no errors, but outputs corrupt lines and not even all the lines.
> Blaming r1:4656 at least gives consistent output and the whole
> file...it looks like r4657 (a one line change to build.conf) causes it
> to flip out. But it doesn't actually throw an error until blaming
> r1:4805 -- after that blame never works for build.conf on Windows (or
> probably any platform where native eol is not LF), it just throws the
> 'inconsistent eols' error.
I've narrowed this down to svn_txdelta_apply_instructions corrupting
the target buffer -- I'm not sure exactly what that means
though...does it mean we have bad deltas in the repository (it works
on Linux so that seems unlikely) or that blame is incorrectly applying
them due to eol size changes or what?
Can someone with more expertise here take a look?
Run 'svn blame -r 4804:4805
http://svn.collab.net/repos/svn/trunk/build.conf' on Windows and the
2nd call to svn_txdelta_apply_instructions will corrupt the target
buffer while processing the window ops.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 2/4/06, D.J. Heap <dj...@gmail.com> wrote:
> On 12/23/05, D.J. Heap <dj...@gmail.com> wrote:
> [snip]
> >
> > I even tried changing the eol-style over a few revisions on a new
> > repository, but still couldn't make it fail. I wonder if the
> > Subversion repository has some mixed eol's in files that have been in
> > it forever or something like that? It works fine on Linux, though, so
> > probably not.
>
>
> I'm looking into this again, and blame seems to act unexpectedly on
> Windows when eol-style is changed. The author of the revision *after*
> the eol-style change gets credit for all lines.
>
> I might have expected that (eol change on every line), but the blame
> code seems to go to lengths to avoid that sort of thing, doesn't it?
>
Even worse, on Windows doing:
svn blame -r1:4657 build.conf
gives no errors, but outputs corrupt lines and not even all the lines.
Blaming r1:4656 at least gives consistent output and the whole
file...it looks like r4657 (a one line change to build.conf) causes it
to flip out. But it doesn't actually throw an error until blaming
r1:4805 -- after that blame never works for build.conf on Windows (or
probably any platform where native eol is not LF), it just throws the
'inconsistent eols' error.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 12/23/05, D.J. Heap <dj...@gmail.com> wrote:
[snip]
>
> I even tried changing the eol-style over a few revisions on a new
> repository, but still couldn't make it fail. I wonder if the
> Subversion repository has some mixed eol's in files that have been in
> it forever or something like that? It works fine on Linux, though, so
> probably not.
I'm looking into this again, and blame seems to act unexpectedly on
Windows when eol-style is changed. The author of the revision *after*
the eol-style change gets credit for all lines.
I might have expected that (eol change on every line), but the blame
code seems to go to lengths to avoid that sort of thing, doesn't it?
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 12/22/05, D.J. Heap <dj...@gmail.com> wrote:
> On 12/22/05, Erik Huelsmann <eh...@gmail.com> wrote:
[snip]
> > PS: If you do have time to spend: there's no blame test failing, which
> > I find alarming. Could you write a test?
Interestingly, I can't make this fail with the tests or with a new
repository. It only seems to fail on files in the Subversion
repository with an eol-style set -- and possibly only on files where
the eol-style changed over blamed revisions.
I even tried changing the eol-style over a few revisions on a new
repository, but still couldn't make it fail. I wonder if the
Subversion repository has some mixed eol's in files that have been in
it forever or something like that? It works fine on Linux, though, so
probably not.
Anyway, if we figure out exactly what is going wrong then I can make a
test for it -- but I'm out of time again for now.
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by "D.J. Heap" <dj...@gmail.com>.
On 12/22/05, Erik Huelsmann <eh...@gmail.com> wrote:
[snip]
>
> If you don't mind me doing it, I'll do it. I've recently changed code
> there and am still working through the resulting conclusions. I'll
> post about it tonight, hopefully resolving the issue over the
> holidays.
That would be great, thanks.
>
>
> bye,
>
> Erik.
>
> PS: If you do have time to spend: there's no blame test failing, which
> I find alarming. Could you write a test?
> Thanks for running the tests and investigating the problems you find!
>
Yes, I can do that though it might be a few days. I wish I had more
time more often to help...
Thanks!
DJ
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Annotate broken on trunk on Win32
Posted by Erik Huelsmann <eh...@gmail.com>.
On 12/22/05, D.J. Heap <dj...@gmail.com> wrote:
> Annotate/blame seems to be broken on trunk on Windows -- the 1.3
> branch works fine.
>
> Here's what I get:
>
> C:\Projects\Subversion>svn annotate
> http://svn.collab.net/repos/svn/trunk/build.conf
> svn: File 'C:\Temp\tmp\tmp.2.tmp' has inconsistent newlines
> svn: Inconsistent line ending style
>
> I'll investigate deeper in a few days when I can.
If you don't mind me doing it, I'll do it. I've recently changed code
there and am still working through the resulting conclusions. I'll
post about it tonight, hopefully resolving the issue over the
holidays.
bye,
Erik.
PS: If you do have time to spend: there's no blame test failing, which
I find alarming. Could you write a test?
Thanks for running the tests and investigating the problems you find!