t5601 breakage at 3cd325f7be (Merge branch 'js/protocol-advertise-multi' into pu, 2018-12-14)

t5601 breakage at 3cd325f7be (Merge branch 'js/protocol-advertise-multi' into pu, 2018-12-14)

[Johannes Schindelin]

[-- Attachment #1: Type: text/plain, Size: 2056 bytes --]

Hi,

this morning Travis sounded quite a few claxons:
https://travis-ci.org/git/git/builds/467839114

It seems that quite a few tests in t5601-clone.sh fail, the first of which
reading like this:

-- snip --
expecting success: 
	git clone myhost:src ssh-clone &&
	expect_ssh "-o SendEnv=GIT_PROTOCOL" myhost src

++ git clone myhost:src ssh-clone
Cloning into 'ssh-clone'...
++ expect_ssh '-o SendEnv=GIT_PROTOCOL' myhost src
++ test_when_finished '
		(cd "$TRASH_DIRECTORY" && rm -f ssh-expect ssh-output.munged && >ssh-output)
	'
++ test 0 = 0
++ test_cleanup='{ 
		(cd "$TRASH_DIRECTORY" && rm -f ssh-expect ssh-output.munged && >ssh-output)
	
		} && (exit "$eval_ret"); eval_ret=$?; :'
++ case "$#" in
++ echo 'ssh: -o SendEnv=GIT_PROTOCOL myhost git-upload-pack '\''src'\'''
++ cd '/Users/vsts/agent/2.144.0/work/1/s/t/trash directory.t5601-clone'
++ sed 's/ssh: -o SendEnv=GIT_PROTOCOL /ssh: /'
++ mv ssh-output.munged ssh-output
++ test_cmp ssh-expect ssh-output
++ diff -u ssh-expect ssh-output
--- ssh-expect	2018-12-14 04:30:28.000000000 +0000
+++ ssh-output	2018-12-14 04:30:28.000000000 +0000
@@ -1 +1 @@
-ssh: -o SendEnv=GIT_PROTOCOL myhost git-upload-pack 'src'
+ssh: myhost git-upload-pack 'src'
error: last command exited with $?=1
not ok 37 - clone myhost:src uses ssh
#	
#		git clone myhost:src ssh-clone &&
#		expect_ssh "-o SendEnv=GIT_PROTOCOL" myhost src
#	
-- snap --

I've bisected this down to 3cd325f7be (Merge branch
'js/protocol-advertise-multi' into pu, 2018-12-14), a merge, meaning that
two topic branches do not play nice with one another.

Staring at the breakage and the changes involved, I suspected that
391985d7c7 (tests: mark & fix tests broken under
GIT_TEST_PROTOCOL_VERSION=1, 2018-12-13) does not play well with the
merged 24c10f7473 (protocol: advertise multiple supported versions,
2018-11-16), and indeed, reverting 391985d7c7 on top of 3cd325f7be lets
t5601 pass again.

It would appear to me, then, that these two patches step on each others'
toes. Josh, Ævar, what should be done about this?

Ciao,
Johannes

Re: t5601 breakage at 3cd325f7be (Merge branch 'js/protocol-advertise-multi' into pu, 2018-12-14)

[Ævar Arnfjörð Bjarmason]

On Fri, Dec 14 2018, Johannes Schindelin wrote:

> Hi,
>
> this morning Travis sounded quite a few claxons:
> https://travis-ci.org/git/git/builds/467839114
>
> It seems that quite a few tests in t5601-clone.sh fail, the first of which
> reading like this:
>
> -- snip --
> expecting success:
> 	git clone myhost:src ssh-clone &&
> 	expect_ssh "-o SendEnv=GIT_PROTOCOL" myhost src
>
> ++ git clone myhost:src ssh-clone
> Cloning into 'ssh-clone'...
> ++ expect_ssh '-o SendEnv=GIT_PROTOCOL' myhost src
> ++ test_when_finished '
> 		(cd "$TRASH_DIRECTORY" && rm -f ssh-expect ssh-output.munged && >ssh-output)
> 	'
> ++ test 0 = 0
> ++ test_cleanup='{
> 		(cd "$TRASH_DIRECTORY" && rm -f ssh-expect ssh-output.munged && >ssh-output)
>
> 		} && (exit "$eval_ret"); eval_ret=$?; :'
> ++ case "$#" in
> ++ echo 'ssh: -o SendEnv=GIT_PROTOCOL myhost git-upload-pack '\''src'\'''
> ++ cd '/Users/vsts/agent/2.144.0/work/1/s/t/trash directory.t5601-clone'
> ++ sed 's/ssh: -o SendEnv=GIT_PROTOCOL /ssh: /'
> ++ mv ssh-output.munged ssh-output
> ++ test_cmp ssh-expect ssh-output
> ++ diff -u ssh-expect ssh-output
> --- ssh-expect	2018-12-14 04:30:28.000000000 +0000
> +++ ssh-output	2018-12-14 04:30:28.000000000 +0000
> @@ -1 +1 @@
> -ssh: -o SendEnv=GIT_PROTOCOL myhost git-upload-pack 'src'
> +ssh: myhost git-upload-pack 'src'
> error: last command exited with $?=1
> not ok 37 - clone myhost:src uses ssh
> #
> #		git clone myhost:src ssh-clone &&
> #		expect_ssh "-o SendEnv=GIT_PROTOCOL" myhost src
> #
> -- snap --
>
> I've bisected this down to 3cd325f7be (Merge branch
> 'js/protocol-advertise-multi' into pu, 2018-12-14), a merge, meaning that
> two topic branches do not play nice with one another.
>
> Staring at the breakage and the changes involved, I suspected that
> 391985d7c7 (tests: mark & fix tests broken under
> GIT_TEST_PROTOCOL_VERSION=1, 2018-12-13) does not play well with the
> merged 24c10f7473 (protocol: advertise multiple supported versions,
> 2018-11-16), and indeed, reverting 391985d7c7 on top of 3cd325f7be lets
> t5601 pass again.
>
> It would appear to me, then, that these two patches step on each others'
> toes. Josh, Ævar, what should be done about this?

Looking at the two the breakage is on my side, but I got away with it
before. I'm re-rolling mine for this & other fixes, and will make sure
the two play well together. Thanks.