Re: [MIT-Scheme-devel] git?

[Chris Hanson]

On Sat, Apr 25, 2009 at 5:16 PM, Matt Birkholz
<address@hidden> wrote:
>> From: Taylor R Campbell <address@hidden>
>> Date: Sat, 25 Apr 2009 19:49:13 -0400
>>
>> [...]
>>
>> I have never tried it.  I have often wanted to have MIT Scheme in a
>> distributed revision control system so that local modifications would
>> be easier to experiment with, but after several attempts to use it I
>> have found Git unusably complicated for me.
>
> git-cvsimport (cvsps) crapped out on me.  I quote:
>
>        Unknown: error
>
> Maybe I will upgrade to Jaunty first...

It won't help.  There is some strangeness in the oldest entries that
defeats git because it only works through the CVS command-line
interface and is unable to deal with the way the revision numbers are
sequenced.  It uses heuristics to reassemble the tree, and the
heuristics break down on these cases.

I tried doing using cvs-fastexport/git-fastimport a couple of months
ago and ran into this problem.  I concluded that it was possible to
make it work but would require some programming to work around the
problems.  The most effective way I thought of was to start from the
RCS files rather than by calling CVS; it's possible to get the RCS
files from savannah for this purpose.  The MIT Scheme RCS parser reads
these entries OK, since the tree structure is explicit in the RCS
files.  It should be pretty simple to write a program to translate
them into git-fastexport form, and use that as input to
git-fastimport.  I haven't done the work to write it -- if you want to
that would be excellent.