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

Re: [PATCH 1/2] fetch: allow passing a transaction to `s_update_ref()`

From
Christian Couder <christian.couder@gmail.com>
Date
Jan 8, 2021, 00:45 UTC
Message-ID
<CAP8UFD1YxHGHLEHd_Bx1awSskjM6fHgM52nPWpTG2UHOmaOT9A@mail.gmail.com>
In-Reply-To
<xmqqtursjqva.fsf@gitster.c.googlers.com>
On Fri, Jan 8, 2021 at 12:04 AM Junio C Hamano <gitster@pobox.com> wrote:
Show 43 quoted lines
>
> Patrick Steinhardt <ps@pks.im> writes:
>
> >  static int s_update_ref(const char *action,
> >                       struct ref *ref,
> > +                     struct ref_transaction *transaction,
> >                       int check_old)
> >  {
> >       char *msg;
> >       char *rla = getenv("GIT_REFLOG_ACTION");
> > -     struct ref_transaction *transaction;
> > +     struct ref_transaction *transaction_to_free = NULL;
> >       struct strbuf err = STRBUF_INIT;
> > -     int ret, df_conflict = 0;
> > +     int ret, df_conflict = 0, commit = 0;
>
> > @@ -597,26 +598,38 @@ static int s_update_ref(const char *action,
> >               rla = default_rla.buf;
> >       msg = xstrfmt("%s: %s", rla, action);
> >
> > -     transaction = ref_transaction_begin(&err);
> > -     if (!transaction ||
> > -         ref_transaction_update(transaction, ref->name,
> > +     /*
> > +      * If no transaction was passed to us, we manage the transaction
> > +      * ourselves. Otherwise, we trust the caller to handle the transaction
> > +      * lifecycle.
> > +      */
> > +     if (!transaction) {
> > +             transaction = transaction_to_free = ref_transaction_begin(&err);
> > +             if (!transaction)
> > +                     goto fail;
> > +             commit = 1;
> > +     }
>
> The idea is that we still allow the caller to pass NULL in which
> case we begin and commit a transaction ourselves, but if the caller
> is to pass an already initialized transaction to us, we obviously
> do not have to "begin" but also we won't commit.
>
> Which makes sense, but then do we still need the "commit" variable
> that reminds us "we are responsible for finishing the transaction we
> started"?

Yeah, I think we also don't need the df_conflict variable, and I don't like the duplication of the following clean up code:

        ref_transaction_free(transaction_to_free);
        strbuf_release(&err);
        free(msg);

Patrick's patch didn't introduce them, but we might still want to get rid of them (see below).

Show 17 quoted lines
> If we rename "transaction_to_free" to a name that makes it more
> clear that it is a transaction that we created and that we are
> responsible for seeing through to its end (after all, "to-free"
> captures only one half of that, i.e. before returning, we are
> responsible for calling ref_transation_free() on it), then we do not
> need such an extra variable that can go out of sync by mistake, no?
> The block that protects the call to ref_transaction_commit() can
> just check if the transaction_to_free variable (with a better name)
> is non-NULL, instead of looking at the commit variable.
>
> Just my attempt to come up with an alternative name for
> transaction_to_free:
>
>  - "our_transaction" because it is ours?
>
>  - "auto_transaction" because it is automatically started and
>    committed without bothering the caller?
"our_transaction" looks fine.
Here is a suggested refactoring patch that could come before Patrick's patch:
-------------------------------------------------------------------------
diff --git a/builtin/fetch.c b/builtin/fetch.c
index ecf8537605..8017fc19da 100644
--- a/builtin/fetch.c
+++ b/builtin/fetch.c
@@ -581,6 +581,22 @@ static struct ref *get_ref_map(struct remote *remote,
#define STORE_REF_ERROR_OTHER 1
#define STORE_REF_ERROR_DF_CONFLICT 2

+static int do_s_update_ref(struct ref_transaction *transaction,
+                          struct ref *ref,
+                          int check_old,
+                          struct strbuf *err,
+                          char *msg)
+{
+       if (!transaction ||
+           ref_transaction_update(transaction, ref->name,
+                                  &ref->new_oid,
+                                  check_old ? &ref->old_oid : NULL,
+                                  0, msg, err))
+               return TRANSACTION_GENERIC_ERROR;
+
+       return ref_transaction_commit(transaction, err);
+}
+
static int s_update_ref(const char *action,
                       struct ref *ref,
                       int check_old)
@@ -589,7 +605,7 @@ static int s_update_ref(const char *action,
       char *rla = getenv("GIT_REFLOG_ACTION");
       struct ref_transaction *transaction;
       struct strbuf err = STRBUF_INIT;
-       int ret, df_conflict = 0;
+       int ret = 0;

       if (dry_run)
               return 0;
@@ -598,30 +614,19 @@ static int s_update_ref(const char *action,
       msg = xstrfmt("%s: %s", rla, action);

       transaction = ref_transaction_begin(&err);
-       if (!transaction ||
-           ref_transaction_update(transaction, ref->name,
-                                  &ref->new_oid,
-                                  check_old ? &ref->old_oid : NULL,
-                                  0, msg, &err))
-               goto fail;

-       ret = ref_transaction_commit(transaction, &err);
+       ret = do_s_update_ref(transaction, ref, check_old, &err, msg);
       if (ret) {
-               df_conflict = (ret == TRANSACTION_NAME_CONFLICT);
-               goto fail;
+               error("%s", err.buf);
+               ret = (ret == TRANSACTION_NAME_CONFLICT)
+                       ? STORE_REF_ERROR_DF_CONFLICT
+                       : STORE_REF_ERROR_OTHER;
       }

       ref_transaction_free(transaction);
       strbuf_release(&err);
       free(msg);
-       return 0;
-fail:
-       ref_transaction_free(transaction);
-       error("%s", err.buf);
-       strbuf_release(&err);
-       free(msg);
-       return df_conflict ? STORE_REF_ERROR_DF_CONFLICT
-                          : STORE_REF_ERROR_OTHER;
+       return ret;
}

static int refcol_width = 10;
-------------------------------------------------------------------------

After the above patch, Patrick's patch would become:

-------------------------------------------------------------------------
diff --git a/builtin/fetch.c b/builtin/fetch.c
index 8017fc19da..d58805c03d 100644
--- a/builtin/fetch.c
+++ b/builtin/fetch.c
@@ -584,6 +584,7 @@ static struct ref *get_ref_map(struct remote *remote,
static int do_s_update_ref(struct ref_transaction *transaction,
                          struct ref *ref,
                          int check_old,
+                          int commit,
                          struct strbuf *err,
                          char *msg)
{
@@ -594,16 +595,17 @@ static int do_s_update_ref(struct
ref_transaction *transaction,
                                  0, msg, err))
               return TRANSACTION_GENERIC_ERROR;

-       return ref_transaction_commit(transaction, err);
+       return (commit) ? ref_transaction_commit(transaction, err) : 0;
}

static int s_update_ref(const char *action,
+                       struct ref_transaction *transaction,
                       struct ref *ref,
                       int check_old)
{
       char *msg;
       char *rla = getenv("GIT_REFLOG_ACTION");
-       struct ref_transaction *transaction;
+       struct ref_transaction *our_transaction = NULL;
       struct strbuf err = STRBUF_INIT;
       int ret = 0;

@@ -613,9 +615,16 @@ static int s_update_ref(const char *action,
               rla = default_rla.buf;
       msg = xstrfmt("%s: %s", rla, action);

-       transaction = ref_transaction_begin(&err);
+       /*
+        * If no transaction was passed to us, we manage the
+        * transaction ourselves. Otherwise, we trust the caller to
+        * handle the transaction lifecycle.
+        */
+       if (!transaction)
+               transaction = our_transaction = ref_transaction_begin(&err);

-       ret = do_s_update_ref(transaction, ref, check_old, &err, msg);
+       ret = do_s_update_ref(transaction, ref, check_old, !!our_transaction,
+                             &err, msg);
       if (ret) {
               error("%s", err.buf);
               ret = (ret == TRANSACTION_NAME_CONFLICT)
@@ -623,7 +632,7 @@ static int s_update_ref(const char *action,
                       : STORE_REF_ERROR_OTHER;
       }

-       ref_transaction_free(transaction);
+       ref_transaction_free(our_transaction);
       strbuf_release(&err);
       free(msg);
       return ret;
...
-------------------------------------------------------------------------

You can find these patches, with Patrick as the author, there:

https://gitlab.com/gitlab-org/gitlab-git/-/commits/cc-improve-s-update-ref/
Previous: Junio C HamanoNext: Junio C Hamano
Message 4 of 41 in “fetch: implement support for atomic reference updates”
  1. 0/2 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 7, 2021
  2. 1/2 fetch: allow passing a transaction to `s_update_ref()`Patrick Steinhardt, Jan 7, 2021
  3. Junio C HamanoJan 7, 2021
  4. Christian CouderJan 8, 2021
  5. Junio C HamanoJan 8, 2021
  6. 2/2 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 7, 2021
  7. Junio C HamanoJan 8, 2021
  8. 0/4 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 8, 2021
  9. 1/4 fetch: extract writing to FETCH_HEADPatrick Steinhardt, Jan 8, 2021
  10. Junio C HamanoJan 8, 2021
  11. Patrick SteinhardtJan 11, 2021
  12. Junio C HamanoJan 11, 2021
  13. 3/4 fetch: allow passing a transaction to `s_update_ref()`Patrick Steinhardt, Jan 8, 2021
  14. Junio C HamanoJan 8, 2021
  15. 2/4 fetch: refactor `s_update_ref` to use common exit pathPatrick Steinhardt, Jan 8, 2021
  16. Junio C HamanoJan 8, 2021
  17. Patrick SteinhardtJan 11, 2021
  18. 4/4 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 8, 2021
  19. Junio C HamanoJan 9, 2021
  20. Patrick SteinhardtJan 11, 2021
  21. Junio C HamanoJan 11, 2021
  22. Patrick SteinhardtJan 12, 2021
  23. Patrick SteinhardtJan 12, 2021
  24. Junio C HamanoJan 12, 2021
  25. 0/5 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 11, 2021
  26. 1/5 fetch: extract writing to FETCH_HEADPatrick Steinhardt, Jan 11, 2021
  27. Junio C HamanoJan 11, 2021
  28. 2/5 fetch: use strbuf to format FETCH_HEAD updatesPatrick Steinhardt, Jan 11, 2021
  29. Patrick SteinhardtJan 11, 2021
  30. Christian CouderJan 11, 2021
  31. Junio C HamanoJan 11, 2021
  32. 4/5 fetch: allow passing a transaction to `s_update_ref()`Patrick Steinhardt, Jan 11, 2021
  33. 3/5 fetch: refactor `s_update_ref` to use common exit pathPatrick Steinhardt, Jan 11, 2021
  34. 5/5 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 11, 2021
  35. Junio C HamanoJan 12, 2021
  36. 0/5 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 12, 2021
  37. 3/5 fetch: refactor `s_update_ref` to use common exit pathPatrick Steinhardt, Jan 12, 2021
  38. 1/5 fetch: extract writing to FETCH_HEADPatrick Steinhardt, Jan 12, 2021
  39. 2/5 fetch: use strbuf to format FETCH_HEAD updatesPatrick Steinhardt, Jan 12, 2021
  40. 4/5 fetch: allow passing a transaction to `s_update_ref()`Patrick Steinhardt, Jan 12, 2021
  41. 5/5 fetch: implement support for atomic reference updatesPatrick Steinhardt, Jan 12, 2021

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.