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

Re: [PATCH v2 2/2] fetch: no redundant error message for atomic fetch

From
Patrick Steinhardt <ps@pks.im>
Date
Dec 15, 2023, 09:56 UTC
Message-ID
<ZXwi0DgfhG4AEE9m@tanuki>
In-Reply-To
<6fb83a00000563a79f3948f9087c634ae507b9f5.1702556642.git.zhiyou.jx@alibaba-inc.com>
On Thu, Dec 14, 2023 at 08:33:12PM +0800, Jiang Xin wrote:
Show 12 quoted lines
> From: Jiang Xin <zhiyou.jx@alibaba-inc.com>
> 
> If an error occurs during an atomic fetch, a redundant error message
> will appear at the end of do_fetch(). It was introduced in b3a804663c
> (fetch: make `--atomic` flag cover backfilling of tags, 2022-02-17).
> 
> In function do_fetch(), a failure message is already shown before the
> retcode is set, so we should not call additional error() at the end of
> this function.
> 
> We can remove the redundant error() function, because we know that
> the function ref_transaction_abort() never fails.

Okay, so this still suffers from the same issue as discussed in the thread at <ZTYue-3gAS1aGXNa@tanuki>, but now it's documented in the commit message. I'm still not convinced that is a good argument to say that the function never fails, and if it ever would it would populate the error message. Especially now where there's churn to introduce the new reftable backend this could change any time.

For the record, I'm proposing to do something like the following:
diff --git a/builtin/fetch.c b/builtin/fetch.c
index fd134ba74d..80b8bc549d 100644
--- a/builtin/fetch.c
+++ b/builtin/fetch.c
@@ -1651,7 +1651,7 @@ static int do_fetch(struct transport *transport,
 	if (atomic_fetch) {
 		transaction = ref_transaction_begin(&err);
 		if (!transaction) {
-			retcode = error("%s", err.buf);
+			retcode = -1;
 			goto cleanup;
 		}
 	}
@@ -1711,7 +1711,6 @@ static int do_fetch(struct transport *transport,
 
 		retcode = ref_transaction_commit(transaction, &err);
 		if (retcode) {
-			error("%s", err.buf);
 			ref_transaction_free(transaction);
 			transaction = NULL;
 			goto cleanup;
@@ -1775,9 +1774,13 @@ static int do_fetch(struct transport *transport,
 	}
 
 cleanup:
+	if (retcode && err.len)
+		error("%s", err.buf);
 	if (retcode && transaction) {
+		strbuf_reset(&err);
 		ref_transaction_abort(transaction, &err);
-		error("%s", err.buf);
+		if (err.len)
+			error("%s", err.buf);
 	}
 
 	display_state_release(&display_state);

This would both fix the issue you observed, but also fixes issues in
case the ref backend failed without writing an error message to the
buffer. It also fixes issues if there were multiple failures, where we'd
print the initial error printed to the buffer twice.

I know this is mostly solidifying us against potential future changes,
but if it's comparatively easy like this I don't see much of a reason
against it.

Patrick

> While we can find a
> common pattern for calling ref_transaction_abort() by running command
> "git grep -A1 ref_transaction_abort", e.g.:
> 
>     if (ref_transaction_abort(transaction, &error))
>         error("abort: %s", error.buf);
> 
> We can fix this issue follow this pattern, and the test case "fetch
> porcelain output (atomic)" in t5574 will also be fixed. If in the future
> we decide that we don't need to check the return value of the function
> ref_transaction_abort(), this change can be fixed along with it.
> 
> Signed-off-by: Jiang Xin <zhiyou.jx@alibaba-inc.com>
> ---
>  builtin/fetch.c         | 4 +---
>  t/t5574-fetch-output.sh | 2 +-
>  2 files changed, 2 insertions(+), 4 deletions(-)
> 
> diff --git a/builtin/fetch.c b/builtin/fetch.c
> index fd134ba74d..01a573cf8d 100644
> --- a/builtin/fetch.c
> +++ b/builtin/fetch.c
> @@ -1775,10 +1775,8 @@ static int do_fetch(struct transport *transport,
>  	}
>  
>  cleanup:
> -	if (retcode && transaction) {
> -		ref_transaction_abort(transaction, &err);
> +	if (retcode && transaction && ref_transaction_abort(transaction, &err))
>  		error("%s", err.buf);
> -	}
>  
>  	display_state_release(&display_state);
>  	close_fetch_head(&fetch_head);
> diff --git a/t/t5574-fetch-output.sh b/t/t5574-fetch-output.sh
> index bc747efefc..8d01e36b3d 100755
> --- a/t/t5574-fetch-output.sh
> +++ b/t/t5574-fetch-output.sh
> @@ -98,7 +98,7 @@ do
>  		opt=
>  		;;
>  	esac
> -	test_expect_failure "fetch porcelain output ${opt:+(atomic)}" '
> +	test_expect_success "fetch porcelain output ${opt:+(atomic)}" '
>  		test_when_finished "rm -rf porcelain" &&
>  
>  		# Clone and pre-seed the repositories. We fetch references into two
> -- 
> 2.41.0.232.g2f6f0bca4f.agit.8.0.4.dev
> 
Previous: Jiang XinNext: Jiang Xin
Message 15 of 19 in “t5574: test porcelain output of atomic fetch”
  1. 1/2 t5574: test porcelain output of atomic fetchJiang Xin, Oct 19, 2023
  2. 2/2 fetch: no redundant error message for atomic fetchJiang Xin, Oct 19, 2023
  3. Patrick SteinhardtOct 23, 2023
  4. Jiang XinOct 23, 2023
  5. Patrick SteinhardtOct 23, 2023
  6. Jiang XinOct 23, 2023
  7. Patrick SteinhardtOct 25, 2023
  8. Junio C HamanoOct 24, 2023
  9. 0/2 jx/fetch-atomic-error-message-fixJiang Xin, Dec 14, 2023
  10. 1/2 t5574: test porcelain output of atomic fetchJiang Xin, Dec 14, 2023
  11. Patrick SteinhardtDec 15, 2023
  12. Jiang XinDec 15, 2023
  13. Junio C HamanoDec 15, 2023
  14. 2/2 fetch: no redundant error message for atomic fetchJiang Xin, Dec 14, 2023
  15. Patrick SteinhardtDec 15, 2023
  16. 0/2 fix fetch atomic error messageJiang Xin, Dec 17, 2023
  17. 1/2 t5574: test porcelain output of atomic fetchJiang Xin, Dec 17, 2023
  18. 2/2 fetch: no redundant error message for atomic fetchJiang Xin, Dec 17, 2023
  19. Patrick SteinhardtDec 18, 2023

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.