Skip to contentSkip to conversation

send-email: restore --in-reply-to superseding behavior

PATCH Superseded Rafael Aquini 3 participants 8 messages Jun 24, 2020 – Jul 1, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 send-email: restore --in-reply-to superseding behavior

git send-email --in-reply-to= fails to override the email headers, if they're present in the output of format-patch, which breakes the contract of the command line switch. This can cause an email thread to break, if subsequent replies to a given message need to be sent after amending fixes to a commit and re-running git format-patch to get the incremental patch-fix posted.

Fixes 256be1d3f0 (send-email: avoid duplicate In-Reply-To/References, 2018-04-17)
---
 git-send-email.perl | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
git-send-email.perl +6 -2
index dc95656f75..342e00b1f3 100755
@@ -1699,10 +1699,14 @@ sub process_file {
 				$xfer_encoding = $1 if not defined $xfer_encoding;
 			}
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to) {
+					$in_reply_to = $1;
+				}
 			}
 			elsif (/^References: (.*)/i) {
-				$references = $1;
+				if (!$initial_in_reply_to) {
+					$references = $1;
+				}
 			}
 			elsif (!/^Date:\s/i && /^[-A-Za-z]+:\s+\S/) {
 				push @xh, $_;
Signature · 1 line
2.23.0
#1 lore raw
Quoted 2 lines · Rafael Aquini

git send-email --in-reply-to= fails to override the email headers, if they're present in the output of format-patch, which breakes the

Will do s/breakes/breaks/ while applying.

It makes me wonder, however, why it is a good idea to have the I-R-T in the format patch output in the first place.

git-send-email.perl sub process_file { · line 1704

Quoted 5 lines
git-send-email.perl +3 -1
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to) {
+					$in_reply_to = $1;
+				}

I can see how this would work the way it should for the first message we send out, so it would work well for a single patch.

But what does this change do to the chaining (either making [PATCH 1/N] thru [PATCH N/N] as responses to the cover letter [PATCH 0/N], or making [PATCH n+1/N] as response to [PATCH n/N] for 1 <= n < N) of multiple messages?

When you prepare a series whose 1..N/N are all pointing at 0/N with the already prepared In-Reply-To (so you have N+1 files to send out), wouldn't you want to make 0/N a reply to a particular message you specify on the command line, while keeping the relationship among your messages intact? Doesn't having $initial_in_reply_to (i.e. command line override) help above code break the chaning?

Quoted 9 lines
git-send-email.perl +3 -1
 			}
 			elsif (/^References: (.*)/i) {
-				$references = $1;
+				if (!$initial_in_reply_to) {
+					$references = $1;
+				}
 			}
 			elsif (!/^Date:\s/i && /^[-A-Za-z]+:\s+\S/) {
 				push @xh, $_;
#2 lore raw
Quoted 7 lines · Junio C Hamano
Quoted 2 lines · Rafael Aquini

git send-email --in-reply-to= fails to override the email headers, if they're present in the output of format-patch, which breakes the

Will do s/breakes/breaks/ while applying.

UGH! I've been fat-fingering typos the whole day, today... Sorry about that one.

Quoted 17 lines

It makes me wonder, however, why it is a good idea to have the I-R-T in the format patch output in the first place.

Review

Quoted 5 lines
(quoted diff) +3 -1
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to) {
+					$in_reply_to = $1;
+				}

I can see how this would work the way it should for the first message we send out, so it would work well for a single patch.

But what does this change do to the chaining (either making [PATCH 1/N] thru [PATCH N/N] as responses to the cover letter [PATCH 0/N], or making [PATCH n+1/N] as response to [PATCH n/N] for 1 <= n < N) of multiple messages?

Quoted 7 lines

When you prepare a series whose 1..N/N are all pointing at 0/N with the already prepared In-Reply-To (so you have N+1 files to send out), wouldn't you want to make 0/N a reply to a particular message you specify on the command line, while keeping the relationship among your messages intact? Doesn't having $initial_in_reply_to (i.e. command line override) help above code break the chaning?

This change will make all emails to appear as a reply to the msgid fed to --in-reply-to. I see your point, though, and at its light I think now this patch, is actually incomplete.

With this change we get back the override desired behavior, but it also breaks the contract, according to the man page.

"
--in-reply-to=<identifier>
Make the first mail (or all the mails with --no-thread) appear as a reply to the given Message-Id, which avoids breaking threads to provide a new patch series. The second and subsequent emails will be sent as replies according to the --[no-]chain-reply-to setting.
"

I drove the change based on my usecase, which is marginal to the multi-part reply case.

I guess we just need the following, for a complete solution:

git-send-email.perl +6 -2
index dc95656f75..768296ea0a 100755
@@ -1699,10 +1699,14 @@ sub process_file {
 				$xfer_encoding = $1 if not defined $xfer_encoding;
 			}
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$in_reply_to = $1;
+				}
 			}
 			elsif (/^References: (.*)/i) {
-				$references = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$references = $1;
+				}
 			}
 			elsif (!/^Date:\s/i && /^[-A-Za-z]+:\s+\S/) {
 				push @xh, $_;
#3 lore raw
Quoted 70 lines · Rafael Aquini
Quoted 7 lines · Junio C Hamano
Quoted 2 lines · Rafael Aquini

git send-email --in-reply-to= fails to override the email headers, if they're present in the output of format-patch, which breakes the

Will do s/breakes/breaks/ while applying.

UGH! I've been fat-fingering typos the whole day, today... Sorry about that one.

Quoted 17 lines

It makes me wonder, however, why it is a good idea to have the I-R-T in the format patch output in the first place.

git-send-email.perl sub process_file { · line 1704

Quoted 5 lines
git-send-email.perl +3 -1
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to) {
+					$in_reply_to = $1;
+				}

I can see how this would work the way it should for the first message we send out, so it would work well for a single patch.

But what does this change do to the chaining (either making [PATCH 1/N] thru [PATCH N/N] as responses to the cover letter [PATCH 0/N], or making [PATCH n+1/N] as response to [PATCH n/N] for 1 <= n < N) of multiple messages?

Quoted 7 lines

When you prepare a series whose 1..N/N are all pointing at 0/N with the already prepared In-Reply-To (so you have N+1 files to send out), wouldn't you want to make 0/N a reply to a particular message you specify on the command line, while keeping the relationship among your messages intact? Doesn't having $initial_in_reply_to (i.e. command line override) help above code break the chaning?

This change will make all emails to appear as a reply to the msgid fed to --in-reply-to. I see your point, though, and at its light I think now this patch, is actually incomplete.

With this change we get back the override desired behavior, but it also breaks the contract, according to the man page.

"
--in-reply-to=<identifier>
Make the first mail (or all the mails with --no-thread) appear as a reply to the given Message-Id, which avoids breaking threads to provide a new patch series. The second and subsequent emails will be sent as replies according to the --[no-]chain-reply-to setting.
"

I drove the change based on my usecase, which is marginal to the multi-part reply case.

I guess we just need the following, for a complete solution:

git-send-email.perl +2 -1
index dc95656f75..768296ea0a 100755
@@ -1699,10 +1699,14 @@ sub process_file {
 				$xfer_encoding = $1 if not defined $xfer_encoding;
 			}
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$in_reply_to = $1;

git-send-email.perl sub process_file { · line 1712

Quoted 10 lines
git-send-email.perl +4 -1
+				}
 			}
 			elsif (/^References: (.*)/i) {
-				$references = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$references = $1;
+				}
 			}
 			elsif (!/^Date:\s/i && /^[-A-Za-z]+:\s+\S/) {
 				push @xh, $_;

This guy worked like a charm, and git send-email, now, follows what the man page says wrt the --in-reply-to usage.

I'll reformat the commit log, and repost the patch ASAP, if you are OK with it.

-- Rafael

#4 lore raw

a test case in t/t9001-send-email.sh will also help, as I am not sure this might be "expected behaviour" as hinted in the man page for git-send-email (under --thread):

"It is up to the user to ensure that no In-Reply-To header already exists
exists when git send-email is asked to add it (especially note that git format-patch can be configured to do the threading itself). Failure to do so may not produce the expected result in the recipient's MUA."

Carlo

#5 lore raw
Quoted 10 lines · Carlo Arenas

a test case in t/t9001-send-email.sh will also help, as I am not sure this might be "expected behaviour" as hinted in the man page for git-send-email (under --thread):

"It is up to the user to ensure that no In-Reply-To header already exists
exists when git send-email is asked to add it (especially note that git format-patch can be configured to do the threading itself). Failure to do so may not produce the expected result in the recipient's MUA."

A quick note: this change does not break that assumption, as well. The "ensure it exists" part means the header must either be in the messages, as populated by format-patch, or it is provided by the --in-reply-to switch option. send-email --thread is not orthogonal, but complementar with --in-reply-to, AFAICS.

The problem we have, right now, is that "send-email --in-reply-to" input gets dropped on the floor if you don't explicitly do "format-patch --no-thread" (or extract a single patch), and this is a behavior regression introduced after v2.17.2

I took a glance in the test-cases, an it seems this is already covered:

test_expect_success $PREREQ 'in-reply-to but no threading' '

        git send-email \
                --dry-run \
                --from="Example <[email protected]>" \
                [email protected] \
                --in-reply-to="<[email protected]>" \
                --no-thread \
                $patches >out &&
        grep "In-Reply-To: <[email protected]>" out

'

Cheers,
-- Rafael

#6 lore raw

PATCH v2 1/1 send-email: restore --in-reply-to superseding behavior

In reply to Junio C Hamano

git send-email --in-reply-to= fails to override In-Reply-To email headers, if they're present in the output of format-patch, even when explicitly told to do so by the option --no-thread, which breaks the contract of the command line switch option, per its man page.

"
--in-reply-to=<identifier>
Make the first mail (or all the mails with --no-thread) appear as a reply to the given Message-Id, which avoids breaking threads to provide a new patch series.
"

This patch fixes the aformentioned issue, by bringing --in-reply-to's old overriding behavior back.

Fixes 256be1d3f0 (send-email: avoid duplicate In-Reply-To/References, 2018-04-17)

---
v2 changelog:
* conform to the command manual page, when send-email threading is requested

 git-send-email.perl | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
git-send-email.perl +6 -2
index dc95656f75..36c47bae1d 100755
@@ -1699,10 +1699,14 @@ sub process_file {
 				$xfer_encoding = $1 if not defined $xfer_encoding;
 			}
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$in_reply_to = $1;
+				}
 			}
 			elsif (/^References: (.*)/i) {
-				$references = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$references = $1;
+				}
 			}
 			elsif (!/^Date:\s/i && /^[-A-Za-z]+:\s+\S/) {
 				push @xh, $_;
Signature · 1 line
2.23.0
#7 lore raw
Quoted 5 lines · Rafael Aquini

This patch fixes the aformentioned issue, by bringing --in-reply-to's old overriding behavior back.

Fixes 256be1d3f0 (send-email: avoid duplicate In-Reply-To/References, 2018-04-17)

the following test case could be squashed on top to make the regression more visible IMHO (it at least pass when applied on top of v2.17.1 and also ra/send-email-in-reply-to-from-command-line-wins)

--- 8< ---

 t/t9001-send-email.sh | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
t/t9001-send-email.sh +13 -1
index 90f61c3400..ec261085ec 100755
@@ -42,7 +42,8 @@ clean_fake_sendmail () {
 }
 
 test_expect_success $PREREQ 'Extract patches' '
-	patches=$(git format-patch -s --cc="One <[email protected]>" [email protected] -n HEAD^1)
+	patches=$(git format-patch -s --cc="One <[email protected]>" [email protected] -n HEAD^1) &&
+	threaded_patches=$(git format-patch -o threaded -s --in-reply-to="format" HEAD^1)
 '
 
 # Test no confirm early to ensure remaining tests will not hang
@@ -1219,6 +1220,17 @@ test_expect_success $PREREQ 'threading but no chain-reply-to' '
 	grep "In-Reply-To: " stdout
 '
 
+test_expect_success $PREREQ 'override in-reply-to if no threading' '
+	git send-email \
+		--dry-run \
+		--from="Example <[email protected]>" \
+		[email protected] \
+		--no-thread \
+		--in-reply-to="override" \
+		$threaded_patches >stdout &&
+	grep "In-Reply-To: <override>" stdout
+'
+
 test_expect_success $PREREQ 'sendemail.to works' '
 	git config --replace-all sendemail.to "Somebody <[email protected]>" &&
 	git send-email \

base-commit: 096547052491426a29e040a5bd94d7f8a4cab8ac --- >8 ---

Quoted 11 lines
git-send-email.perl +2 -1
index dc95656f75..36c47bae1d 100755
@@ -1699,10 +1699,14 @@ sub process_file {
 				$xfer_encoding = $1 if not defined $xfer_encoding;
 			}
 			elsif (/^In-Reply-To: (.*)/i) {
-				$in_reply_to = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$in_reply_to = $1;

git-send-email.perl sub process_file { · line 1712

Quoted 10 lines
git-send-email.perl +4 -1
+				}
 			}
 			elsif (/^References: (.*)/i) {
-				$references = $1;
+				if (!$initial_in_reply_to || $thread) {
+					$references = $1;
+				}
 			}
 			elsif (!/^Date:\s/i && /^[-A-Za-z]+:\s+\S/) {
 				push @xh, $_;

in both cases, doing !defined $initial_in_reply_to might seem like a more consistent option.

Carlo