git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH v3 03/11] promisor-remote: implement promisor_remote_get_direct()

From
Christian Couder <christian.couder@gmail.com>
Date
Apr 1, 2019, 16:41 UTC
Message-ID
<CAP8UFD1SK2mFBYJLqY9VE_Z5tQ5Mor2guc-X3yTbzfUZbHM_2Q@mail.gmail.com>
In-Reply-To
<xmqqpnqve71d.fsf@gitster-ct.c.googlers.com>
On Wed, Mar 13, 2019 at 5:24 AM Junio C Hamano <gitster@pobox.com> wrote:
>
> Christian Couder <christian.couder@gmail.com> writes:
Show 17 quoted lines
> > +int promisor_remote_get_direct(const struct object_id *oids, int oid_nr)
> > +{
> > +     struct promisor_remote *o;
> > +
> > +     promisor_remote_init();
> > +
> > +     for (o = promisors; o; o = o->next) {
> > +             if (fetch_objects(o->remote_name, oids, oid_nr) < 0)
> > +                     continue;
> > +             return 0;
>
> Suppose the caller asks to fetch 3 objects, A, B and C, from two
> promisors.  The first promisor can give you A and B but cannot give
> you C.  The second promisor only can give you C.
>
> Does fetch_objects() return failure after attempting to fetch A, B
> and C from the first promisor?
Yes, I think it should still send the objects it has and then fail.

Maybe it doesn't work like this right now (I haven't checked), or maybe the failure should be different than a regular failure. In those cases I can maybe improve on that in a later iteration.

Show 5 quoted lines
> Then we go on to the second promisor
> but do we ask all three?  That would mean the second promisor will
> also fail because it cannot give you A and B, and then the whole
> thing would fail.  It would be nicer if the mechanism would allow us
> to fetch what is still missing from later promisor, perhaps.
Yeah, I implemented fetching only what is still missing in v4.
Show 5 quoted lines
> As the original "fetch" protocol only allows you to fetch a pack
> with everything you asked for in it, instead of feeding you a pack
> with best effort, I think the answer to the above is "it is very
> hard to improve over what we have here", but people may have
> interesting ideas ;-)

I think we should make it best effort when acting as a promisor remote. I will take a look at that.

Show 7 quoted lines
> > +     }
> > +
> > +     return -1;
> > +}
> > +
>
> Adding trailing blank line at the end?
Removed.
Previous: Junio C HamanoNext: Christian Couder
Message 5 of 22 in “Many promisor remotes”
  1. 00/11 Many promisor remotesChristian Couder, Mar 12, 2019
  2. 01/11 fetch-object: make functions return an error codeChristian Couder, Mar 12, 2019
  3. 03/11 promisor-remote: implement promisor_remote_get_direct()Christian Couder, Mar 12, 2019
  4. Junio C HamanoMar 13, 2019
  5. Christian CouderApr 1, 2019
  6. 02/11 Add initial support for many promisor remotesChristian Couder, Mar 12, 2019
  7. Junio C HamanoMar 13, 2019
  8. Junio C HamanoMar 13, 2019
  9. Christian CouderApr 1, 2019
  10. 04/11 promisor-remote: add promisor_remote_reinit()Christian Couder, Mar 12, 2019
  11. Junio C HamanoMar 13, 2019
  12. Christian CouderApr 1, 2019
  13. 05/11 promisor-remote: use repository_format_partial_cloneChristian Couder, Mar 12, 2019
  14. Junio C HamanoMar 13, 2019
  15. Christian CouderApr 1, 2019
  16. Junio C HamanoApr 1, 2019
  17. 07/11 promisor-remote: parse remote.*.partialclonefilterChristian Couder, Mar 12, 2019
  18. 08/11 builtin/fetch: remove unique promisor remote limitationChristian Couder, Mar 12, 2019
  19. 06/11 Use promisor_remote_get_direct() and has_promisor_remote()Christian Couder, Mar 12, 2019
  20. 09/11 t0410: test fetching from many promisor remotesChristian Couder, Mar 12, 2019
  21. 10/11 partial-clone: add multiple remotes in the docChristian Couder, Mar 12, 2019
  22. 11/11 remote: add promisor and partial clone config to the docChristian Couder, Mar 12, 2019

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.