Subversion revisions references were not ported

psf / gh-migration

This repo is used to manage the migration from bugs.python.org to GitHub.

42 stars 8 forks source link

Subversion revisions references were not ported #23

Open serhiy-storchaka opened 2 years ago

serhiy-storchaka commented 2 years ago

For example:

https://bugs.python.org/issue2116#msg87822

Committed in r72662, r72670. Thanks!

https://github.com/python/cpython/issues/46370#issuecomment-1093409709

Committed in r72662, r72670. Thanks!

ezio-melotti commented 2 years ago

Those refs were already broken since they mapped to hg.python.org, which hasn't been accessible in a while, so I didn't port them.

gvanrossum commented 2 years ago

I assume you meant svn.python.org? hg.python.org is still alive.

ezio-melotti commented 2 years ago

The rXXXXX revisions were updated in bpo to point to hg.python.org/lookup, so that e.g. r72662 linked to https://hg.python.org/lookup/r72662. Then the lookup script was supposed to find the hg changeset equivalent to r72662, however if I try to open that link I get this:

The title of the popup indeed mentions svn.python.org, but the end result is the same: the link and the lookup script don't seem accessible anymore.

gvanrossum commented 2 years ago

Hm, it looks like it might be difficult to recover those references. E.g. I looked for the commit that added weakref support to copy.py, which was r72662 according to https://bugs.python.org/issue2116#msg87822 (or r72670?). It's easy to locate in the git history for copy.py: https://github.com/python/cpython/commit/6e61006cc2dfa6c07fa84126622685769bac635d -- but somehow the revision number got changed:

Merged revisions 72669 via svnmerge from
svn+ssh://pythondev@svn.python.org/python/trunk

........
  r72669 | antoine.pitrou | 2009-05-15 18:54:52 +0200 (ven., 15 mai 2009) | 3 lines

  Issue https://github.com/python/cpython/pull/2116: Weak references and weak dictionaries now support copy()ing and deepcopy()ing.
........

serhiy-storchaka commented 2 years ago

Yes, it was already broken on BPO for some time. It is bad. We are losing history. And it is only 13 years old.

I think it is a good opportunity to fix old references. Can we scan all logs and create a mapping between Subversion revision numbers, and Mercurial and Git hashes?

ezio-melotti commented 2 years ago

I found two files used by the lookup script:

hglookup.py is the script itself
hg_commits.json is a flat list of HG commits

There doesn't seem to be any mapping between SVN and HG, and the script simply accepts rXXXXX and redirects that to svn.python.org (see line 48).

I'm not aware of a mapping between HG and Git either, unless someone still has the logs of the SVN->HG and HG->Git migrations. It should be possible to recreate them though -- assuming we still have an SVN clone around (or if the SVN commit number is stored somewhere in the HG metadata).