= Notes/Subversion/Importing =

== Importing RCS files to a repository ==

STILL WORKING ON THIS PAGE

=== Initial Import ===
Just importing the miriad tree as it stands is unsatisfactory, because we lose the revision history. So we use {{{cvs2svn}}}.

'''Note:''' there is a [http://cvs2svn.tigris.org/issues/buglist.cgi?target_milestone=---&issue_status=UNCONFIRMED&issue_status=NEW&issue_status=STARTED&issue_status=REOPENED&issue_status=RESOLVED&issue_status=VERIFIED&issue_status=CLOSED&field0-0-0=longdesc&type0-0-0=substring&value0-0-0=file%20already%20exists&field0-0-1=short_desc&type0-0-1=substring&value0-0-1=file%20already%20exists&field0-0-2=status_whiteboard&type0-0-2=substring&value0-0-2=file%20already%20exists bug] in the version of {{{cvs2svn}}} available in the Debian ''sarge'' release that causes the import to die partway through. The version available in Debian ''etch'' works correctly.

{{{
 % mkdir miriad
 % rsync -av /nfs/atapplic/miriad/ miriad/

 # method 1
 # we'll create a new repository in a directory named 'svnmiriad'
 % cvs2svn -s svnmiriad  miriad

 # method 2
 # we can create an empty repository named 'svnmiriad' and load via a dumpfile
 % cvs2svn --dumpfile=20071123.svndump miriad
 % svnadmin load svnmiriad < 20071123.svndump

 # method 3
 # add to an existing repository
 % svnadmin create svnmiriad
 % cvs2svn --existing-repos -s svnmiriad miriad

}}}

To do later imports, we would have to do something like this:
{{{
  % mkdir miriad2
  % rsync -a /nfs/atapplic/miriad/ miriad2/
  # create a new repos
  % cvs2svn -s svnmiriad2 miriad2
  # assuming a repository named 'svnmiriad'
  % svn merge -rM:N svnmiriad2 svnmiriad  #???
  # M and N are revision numbers in the svnmiriad2 repository.
}}}

Not everything in the {{{miriad}}} directory gets imported (e.g. the binaries), because they have no RCS file.

=== next phase ===
tag the raw import, before we start to muck around with it.

need to remove the compiled binaries
  No - the binaries never get checked into RCS so they aren't imported to svn.
need to check out the code (and svn add all the files?)
  No - just do the cvs2svn dump. that does the checkout but it leaves
  the source in the '.rcs' directory. Must move it from there.
  The dump leaves darwin_ppc, darwin_x86 there as well.
  These contain just a short GNUmakedefs file. Diffing them against the version in {{{scripts/darwin_*}}} indicates they are cruft and can be removed.

check for things that should have been in RCS and weren't.
  This reveals some files with {{{,x}}} as their RCS suffix instead of the default {{{,v}}}.
  {{{etc/archive}}} is not (all) in RCS.  Ditto {{{linpack, config, at_friends}}}.

  {{{linux, linux64, sun4sol, unicos}}} platform-specific directories aren't there but the {{{darwin_}}} ones are. 

check the programs are the same at the current revision.



need to change the chkout, chklog targets
need to add handlers of .svn directories

The GNUmakedefs defines a 'MIRFTP' variable; the comment about solaris-only is wrong.

----

[wiki:Notes Notes Home]