On Sun Sep 22, 2024 at 00:19, Bence Ferdinandy <bence@xxxxxxxxxxxxxx> wrote: > > On Sat Sep 21, 2024 at 15:40, Phillip Wood <phillip.wood123@xxxxxxxxx> wrote: >> On 19/09/2024 13:13, Bence Ferdinandy wrote: >> > Add a new REF_CREATE_ONLY flag for use by the files backend which will >> > only update the symref if it doesn't already exist. Add the possibility >> > to pass extra flags to refs_update_symref so that it can utilize this >> > new flag. >> >> I'm not sure we need a new flag to do this as it is already supported by >> the ref transaction api. > > Thanks, I was not aware of ref_transaction_create. It also seems to return with > TRANSACTION_NAME_CONFLICT so we should be able to see from the error code if > indeed the existence was the problem or something else went wrong. Unfortunately, it seems that my reading of the code did not pass practice. When using ref_transaction_create ref_transaction_commit will return with -2 if the reference already exists, but it also returns with -2 for various other issues, like if the lock file already exists. I could parse the error message to see what was the cause, but that doesn't feel like a robust solution. Since fetch should _not_ error out on this, I think the REF_CREATE_ONLY flag is warranted. As it stands, it would serve a different purpose than ref_transaction_create, i.e. a "silent" create-only. I'll send a v4 tomorrow hopefully. Best, Bence
