vcs-fast-import-devs team mailing list archive

[Sverre Rabbelier <srabbelier@xxxxxxxxx>]

[+vcs-fast-import-devs, not culled for their benefit]

On Sun, Aug 29, 2010 at 13:59, Daniel Barkalow <barkalow@xxxxxxxxxxxx> wrote:
> On Sat, 28 Aug 2010, Sverre Rabbelier wrote:
>> Currently the only way to end an import stream is to close it, which
>> is not desirable when the stream that's being used is shared. For
>> example, the remote helper infrastructure uses a pipe between it and
>> the helper process, part of the protocol is to send a fast-import
>> stream accross. Without a way to end the stream the remote helper
>> infrastructure is forced to limit itself to have a command that uses
>> a fast-import stream as it's last command.
>>
>> Add a trivial 'done' command that causes fast-import to stop reading
>> from the stream and exit.
>
> Yeah, this is definitely worthwhile.
>
>> ---
>>
>>   Very straightforward. It is handled in parse_feature() instead of
>>   in parse_one_feature() because I didn't want to allow '--done' as a
>>   commandline argument. Allowing it would be silly, it surves no
>>   other purpose than to indicate up front that the stream will
>>   contain a 'done' command at the end.
>>
>>   I'm fine too with dropping the feature and just adding the new
>>   command, whichever is preferred.
>
> I think the point of the feature would be to get the error response up
> front, where it might be easier to determine what to do about importers
> not supporting it. As such, I think the command line option actually makes
> at least as much sense, but it's probably not necessary anyway.
>
> I believe there's a gfi mailing list, which ought to hear about this bit.

I've added them.

> Not that there are likely to be conflicts, but, when I was thinking about
> adding this command (for the same reason you're adding it), I'd called it
> "quit", so it's worth letting people know a de facto standard, so gfi
> implementations don't vary.

Agreed, I've added it to the fastimport python library without much trouble

> The code looks obviously good to me.

Thanks.

-- 
Cheers,

Sverre Rabbelier