Kousik Sanagavarapu <five231003@xxxxxxxxx> writes: > A collision test is triggered in sha1_object(), whenever there is an > object file in our repo. If our repo is a partial clone, then checking > for this file existence does not lazy-fetch the object (if the object > is missing and if there are one or more promisor remotes) when > fetch_if_missing is set to 0. > ... > Hence, use has_object() to check for the existence of an object, which > has the default behavior of not lazy-fetching in a partial clone. It is > worth mentioning that this is the only place where there is potential for > lazy-fetching and all other cases [2] are properly handled, making it safe > to remove this global here. > > [1] See 8b4c0103a9 (sha1_file: support lazily fetching missing objects, > 2017-12-08) Thanks for this reference. The way I read the "lazy fetching is by default not suppressed, and this is a temporary measure" described in the log message is quite opposite from where this patch wants to go, though. I think the commit envisioned the world where all the accesses to the object layer are aware of the characteristics of a lazy clone (e.g. has_object() reporting "we do not have that object locally (yet)" is not an immediate sign of a repository corruption) and when to lazily fetch and when to tolerate locally missing objects is controlled more tightly, and to reach that world one step at a time, introduced the global, so that for now everybody lazily fetches, but the commands individual patches concentrates to "fix" can turn off the "by default all missing objects are lazily fetched" so that they can either allow certain objects to be locally missing, or fetch them from promisor remotes when they need the contents of such objects. By fixing each commands one by one, eventually we would be able to wean ourselves away from this "by default everything is lazily fetched" global---in other words, in the ideal endgame, the fetch_if_missing should be set to 0 everywhere. So, if this patch was made in reaction to the "it was a temporary measure" in 8b4c0103 (sha1_file: support lazily fetching missing objects, 2017-12-08), I think it goes in the completely opposite direction. If the patch shared the cause with 8b4c0103 and wanted to help realize the ideal world, it instead should have left this command who can already work with fetch_if_missing=0 alone and fixed somebody else who still depends on fetch_if_missing=1, I think. Now it is a separate issue to argue if "everybody knows exactly when to trigger lazy fetching and fetch_if_missing is set to false everywhere" is really the ideal endgame. I do not think "Future patches will update some commands to either tolerate missing objects or be more efficient in fetching them." proposed by the commit from late 2017 has seen that much advance recently. But for commands that need to deal with many missing objects, enumerating the objects that are missing locally and need fetching first and then requesting them in a batch should be vastly more efficient than the default lazy fetch logic that lets the caller request a single object, realize it is missing locally, make a connection to fetch that single object and disconnect. So I have to suspect that ... > This global was added as a temporary measure to suppress the fetching > of missing objects [1] and can be removed once the remaining commands: > - fetch-pack > - fsck > - pack-objects > - prune > - rev-list > can handle lazy-fetching without fetch_if_missing. ... this "can handle" may be a misguided direction to go in. They were taught not to lazy fetch because blindly lazy fetching was bad, weren't they?
