Re: [PATCH v2 3/4] t5300: move --window clamp test next to unclamped

> Jeff King <peff@peff.net> writes:
> 
> >> As far as I know, index-pack, when run as part of fetch, indexes a pack
> >> that's not in the repository's object store; it indexes a packfile in a
> >> temp directory. (So I don't think this is a strange thing to do.)
> >
> > When fetching (or receiving a push), we use "index-pack --stdin" and do
> > write the resulting pack into the repository (and the command will
> > complain if there is no repository).
> > ...
> >> We definitely should prevent the segfault, but I think that's better
> >> done by making --promisor only work if we run index-pack from within a
> >> repo. I don't think we can restrict the repacking to run only if we're
> >> indexing a pack within the repo, because in our fetch case, we're
> >> indexing a new pack - not one within the repo.
> >
> > I think the "--stdin" thing above neatly solves this.
> [...]
> 
> Tying this extra processing to the use of "--stdin" is not exactly
> intuitive, in that a "--stdin" user is not necessarily doing a fetch
> (even though a fetch may always use "--stdin"), but I guess it is a
> good enough approximation (and the best one easily available to us)
> if we want to safeguard the use of this "--promisor" logic only to
> fetch client.