--use-external-blob-generator (Re: Handling of warnings from cvs)

Michael Haggerty wrote:

>             I wonder whether it would be worth the extra work to enable
> it to work with a non-seekable output?  I'd assumed that git-fast-import
> is so vastly faster than the intervening cvs2git steps that the
> parallelism wouldn't gain much in wall-clock time.  (Of course it would
> save having to write the blobs to a big temporary file.)

That last bit (hitting disk) is half of what motivated me to try it.

>> I can avoid that by redirecting output to a seekable file, but then
>> fast-import crashes.
>
> It sounds like this implies that the --use-external-blob-generator
> option plus git-fast-import is broken no matter how you use it, but
> below you propose removing its "experimental" designation.  So I must be
> misunderstanding something.  Is the problem that the blob and dump data
> are interleaved?

Exactly.  The dump data refers to blobs that have not been dumped yet.
("Crashes" was not too precise a term: the malformed stream is
rejected rather than causing a segfault.)

>                   If so, then it shouldn't be too hard to have cvs2git
> wait until the external blob generator is finished before starting to
> generate dumpfile info.

That would definitely address the other half of my motivation ---
user-friendliness.  I'm a fan of 'fire and forget' conversion tools,
like this:

 cvs2git /path/to/cvs/root --stdout | git fast-import

Or like this:

 cvs2git /path/to/cvs/root --fast-import="git fast-import"

> Thanks for the patch.  Committed as r5304 (followed by some related
> changes).

Looks good to me.

>>  There is a "git fast-import" patch in flight that would allow
>>  removing that second requirement by asking fast-import for a
>>  reminder about blobs past.
[...]
> I've been half-following that patch on the git mailing list but haven't
> yet seriously thought about using the new feature due to lack of time
> plus the rationalization listed above :-)

I might take a stab at it if time allows.

Regards,
Jonathan

------------------------------------------------------
http://cvs2svn.tigris.org/ds/viewMessage.do?dsForumId=1667&dsMessageId=2682119

To unsubscribe from this discussion, e-mail: [[email protected]].