bzr update exceeds pypy's maximum recursion depth in gigantic repositories such as MariaDB.

Apparently this is a real infinite loop bug, because increasing the maximum stack size by adding "sys.setrecursionlimit(10000)" to the bzr python script actually still triggers this bug. My now much larger .bzr.log file is attached.

Also, setting it much higher than 10000 results in a segmentation fault, because the stack overflows into the heap.

My computer is too old to run the binary compiled for Ubuntu on my Slackware setup, so I'm stuck with Cpython or pypy 1.9. I don't have enough RAM to compile pypy myself, because a 64bit build needs 4gig, and I only have 4gigs of total ram. pypy's build requirements are insane.

Thanks,
Dave.

The actual traceback indicates you have a branch that is pointing at itself. You might try looking at ".bzr/branch/location" information for a location that ends up pointing in a loop.

That's kinda funny, because it's MariaDB's open source repo. I have made no local commits or changes. The only bzr commands I've run are ones to list tags, and to try to change to a different tag.

The only thing in ".bzr/branch/location" is:

file:///home/dly/Desktop/Code/maria/trunk/

Which is the local path to the repository. Does having this there create a loop? I deleted the file, and that caused me to get an error about the file not existing. Then, I created an empty file with the correct name using touch. I no longer got the error message, but I still got the crazy back trace. So, if there is a loop, it can't be because of the contents of that file, because it crashes with a back trace even when that file is empty. A new .bzr.log file has been attached. The latest stuff will be the last few bzr tags and bzr update runs.

I also created an archive of the MariaDB repo's .bzr directory. It's only 5.3 megabytes. The .bzr/branch/location.old file is the old .bzr/branch/location file you can rename it to that if you want.

Thanks,
Dave.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 2013-10-03 8:26, David Yingling wrote:
> That's kinda funny, because it's MariaDB's open source repo. I have
> made no local commits or changes. The only bzr commands I've run
> are ones to list tags, and to try to change to a different tag.
>
> The only thing in ".bzr/branch/location" is:
>
> file:///home/dly/Desktop/Code/maria/trunk/
>
> Which is the local path to the repository. Does having this there
> create a loop? I deleted the file, and that caused me to get an
> error about the file not existing. Then, I created an empty file
> with the correct name using touch. I no longer got the error
> message, but I still got the crazy back trace. So, if there is a
> loop, it can't be because of the contents of that file, because it
> crashes with a back trace even when that file is empty. A new
> .bzr.log file has been attached. The latest stuff will be the last
> few bzr tags and bzr update runs.

"The local path to the repository" means it is pointing at itself. If
you make it an empty file, I'm guessing we interpret "" using the
local directory, and again it points back at itself.

I don't know how you ended up checking out a branch to itself. (though
the freedom with which you edit files under .bzr/ is a bit concerning)
What I would have expected is to just have a regular branch there.

>
> I also created an archive of the MariaDB repo's .bzr directory.
> It's only 5.3 megabytes. The .bzr/branch/location.old file is the
> old .bzr/branch/location file you can rename it to that if you
> want.
>
> Thanks, Dave.
>

So (a) having a .bzr/branch/location file means that you don't have a
local branch, but you have a local working tree pointing at "some
other branch". In this case, that somehow ended up pointed directly
back at itself. So you had a lightweight checkout pointing at a branch
that is pointing at itself and we try to open it, to see that it is a
reference and we open that, etc.

It needs to point "somewhere else". Probably the easiest way to do
that is with "bzr switch --force" which would be something like:

bzr branch $MARIADB /home/dly/Desktop/Code/maria/alt-trunk
cd /home/dly/Desktop/Code/maria/trunk
bzr switch --force home/dly/Desktop/Code/maria/alt-trunk

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.13 (Cygwin)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iEYEARECAAYFAlJNNnkACgkQJdeBCYSNAAM07ACgk8SZQgg62LHaWbQuhpLoT6AT
wqoAn0cGjYntHQxEHZaD3yvSvbuCpPYq
=F40Y
-----END PGP SIGNATURE-----

Wow, I can't believe it was a lightweight checkout. It really did take 1.2gigs or so of ram and like 45 minutes or so to download what is basically the same thing in a 45 meg MariaDB release + 20 megs of state for the .bzr directory.

I tried your suggestion, but it failed, because the first bzr branch command also fails with a python "maximum recursion depth exceeded" error message. The back trace is included in the latest .bzr.log file I've attached.
Apparently bzr can't really do anything when its location file is messed up.

So, is the bug fix just an error message for when the location file is messed up? I think an error message would be better than a python exception from an infinite recursion loop.

The other thing I tried was the MariaDB repo that is available for download. It's mentioned on the page https://mariadb.com/kb/en/getting-the-mariadb-source-code/ . You just download and unarchive it. However, it is just a bzr repo with no branches checked out, so you need to do a "bzr branch lp:maria" to get an actual bzr checkout, because most commands don't seem to work when nothing is checked out such as "bzr tags."

When running the bzr branch lp:maria command it seems to connect to lanuch pad, and download some stuff, and then it crashes with a segmentation fault:

dly@betty mariadb2> bzr branch lp:maria/5.5
You have not informed bzr of your Launchpad ID, and you must do this to
write to Launchpad or access private data. See "bzr help launchpad-login".
Segmentation fault -

The "Segmentation fault - " replaces where the network activity status was. A back trace of this error should also be in the attached .bzr.log. So using that repo doesn't work for me either.

I tried reinstalling bzr a few times. Each time I get the "Cannot build extension "bzrlib._annotator_pyx"." error. So, I do the "build_ext --allow-python-fallback" trick to get it to build anyway. I'm starting to doubt my bug is a real bug in bazaar, but instead a bug in my install of bzr, which seems to be messed up. I'm running an older Slackware 13.37 release, which is a few years old, which may have contributed to these issues.

So, could you please download the MariaDB repo as explained on the page https://mariadb.com/kb/en/getting-the-mariadb-source-code/ in the section "Source Tree Tarball." It's really simple just download the file. Remember to make a new directory and cd to it, because unarchiving the archive will create a .bzr directory in the current directory not in a containing directory. Then supposedly running the command "bzr branch lp:maria" will then checkout the latest version. I get a segmentation fault when I do this, but you shouldn't.

Anyway, after doing all that to setup a MariaDB repo just type the command "bzr tags." It will list a whole bunch of tags. Pick one representing a newish MariaDB release. I forget their exact naming--something like: "Maria-5.5" "MariaDB-10.0" or something with a 5.5 or 10.0 version number, which are the latest releases. Then try to checkout that tag using s...