"./t0001-init.sh --valgrind" is broken

Hi,

It looks like commit 57ea7123c86771f47f34e7d92d1822d8b429897a (git.c:
make sure we do not leak GIT_* to alias scripts, Dec 20 14:50:19 2015)
broke "./t0001-init.sh --valgrind".

I get:

expecting success:
        (
                env | sed -ne "/^GIT_/s/=.*//p" &&
                echo GIT_PREFIX &&        # setup.c
                echo GIT_TEXTDOMAINDIR    # wrapper-for-bin.sh
        ) | sort | uniq >expected &&
        cat <<-\EOF >script &&
        #!/bin/sh
        env | sed -ne "/^GIT_/s/=.*//p" | sort >actual
        exit 0
        EOF
        chmod 755 script &&
        git config alias.script \!./script &&
        ( mkdir sub && cd sub && git script ) &&
        test_cmp expected actual

--- expected    2016-03-03 00:05:17.113754381 +0000
+++ actual      2016-03-03 00:05:19.041783583 +0000

@@ -10,7 +10,6 @@
 GIT_PREFIX
 GIT_TEMPLATE_DIR
 GIT_TEST_TEE_STARTED
-GIT_TEXTDOMAINDIR
 GIT_TRACE_BARE
 GIT_VALGRIND
 GIT_VALGRIND_ENABLED

not ok 6 - No extra GIT_* on alias scripts

It's late here so I don't have time to work on this tonight.

Thanks,
Christian.

On Thu, Mar 3, 2016 at 7:07 AM, Christian Couder
[off-list ref] wrote:

Hi,

It looks like commit 57ea7123c86771f47f34e7d92d1822d8b429897a (git.c:
make sure we do not leak GIT_* to alias scripts, Dec 20 14:50:19 2015)
broke "./t0001-init.sh --valgrind".

Just wanted to confirm the problem. I will look at it later today.
-- 
Duy

Am 03.03.2016 um 02:04 schrieb Duy Nguyen:

On Thu, Mar 3, 2016 at 7:07 AM, Christian Couder
[off-list ref] wrote:

quoted

Hi,

It looks like commit 57ea7123c86771f47f34e7d92d1822d8b429897a (git.c:
make sure we do not leak GIT_* to alias scripts, Dec 20 14:50:19 2015)
broke "./t0001-init.sh --valgrind".

Just wanted to confirm the problem. I will look at it later today.

Here's a patch.

---- 8< ----
Subject: [PATCH] t0001: fix GIT_* environment variable check under --valgrind

When a test case is run without --valgrind, the wrap-for-bin.sh
helper script inserts the environment variable GIT_TEXTDOMAINDIR, but
when run with --valgrind, the variable is missing. A recently
introduced test case expects the presence of the variable, though, and
fails under --valgrind.

Rewrite the test case to strip conditially defined environment variables
from both expected and actual output.

Signed-off-by: Johannes Sixt <redacted>
---
 t/t0001-init.sh | 20 +++++++++-----------
 1 file changed, 9 insertions(+), 11 deletions(-)

diff --git a/t/t0001-init.sh b/t/t0001-init.sh
index 295aa59..a5b9e7a 100755
--- a/t/t0001-init.sh
+++ b/t/t0001-init.sh

@@ -88,19 +88,17 @@ test_expect_success 'plain nested in bare through aliased command' '
 '
 
 test_expect_success 'No extra GIT_* on alias scripts' '
-	(
-		env | sed -ne "/^GIT_/s/=.*//p" &&
-		echo GIT_PREFIX &&        # setup.c
-		echo GIT_TEXTDOMAINDIR    # wrapper-for-bin.sh
-	) | sort | uniq >expected &&
-	cat <<-\EOF >script &&
-	#!/bin/sh
-	env | sed -ne "/^GIT_/s/=.*//p" | sort >actual
-	exit 0
+	write_script script <<-\EOF &&
+	env |
+		sed -n \
+			-e "/^GIT_PREFIX=/d" \
+			-e "/^GIT_TEXTDOMAINDIR=/d" \
+			-e "/^GIT_/s/=.*//p" |
+		sort
 	EOF
-	chmod 755 script &&
+	./script >expected &&
 	git config alias.script \!./script &&
-	( mkdir sub && cd sub && git script ) &&
+	( mkdir sub && cd sub && git script >../actual ) &&
 	test_cmp expected actual
 '
 

-- 
2.7.0.118.g90056ae

+the-other-Johannes who added valgrind support.

On Thu, Mar 3, 2016 at 1:55 PM, Johannes Sixt [off-list ref] wrote:

---- 8< ----
Subject: [PATCH] t0001: fix GIT_* environment variable check under --valgrind

When a test case is run without --valgrind, the wrap-for-bin.sh
helper script inserts the environment variable GIT_TEXTDOMAINDIR, but
when run with --valgrind, the variable is missing. A recently
introduced test case expects the presence of the variable, though, and
fails under --valgrind.

Yep.

It's interesting though that valgrind sets up some variables without
going through bin-wrappers. That's understandable because valgrind
support is added (in 4e1be63) 10 months before bin-wrappers (in
ea92519).  But it's probably better that we inject valgrind command
from inside bin-wrappers script, the same way we inject gdb, I think.

Rewrite the test case to strip conditially defined environment variables
from both expected and actual output.

Or we could set GIT_TEXTDOMAINDIR in the "if test -n $valgrind" code
in test-lib.sh, which makes the two more consistent. Also simpler
patch.

-- 
Duy

On Thu, Mar 3, 2016 at 7:09 PM, Duy Nguyen [off-list ref] wrote:

But it's probably better that we inject valgrind command
from inside bin-wrappers script, the same way we inject gdb, I think.

For the best of both worlds, we should recreate bin-wrappers in
test-lib.sh (i.e. the valgrind way), not in Makefile. Somewhat
unrelated, but because topdir is getting really crowded and
bin-wrappers is used for the test suite only, it should be moved
inside t/ (i'm going to move all test-* to t/ too, later).
-- 
Duy

On Thu, Mar 03, 2016 at 07:09:12PM +0700, Duy Nguyen wrote:

+the-other-Johannes who added valgrind support.

On Thu, Mar 3, 2016 at 1:55 PM, Johannes Sixt [off-list ref] wrote:

quoted

---- 8< ----
Subject: [PATCH] t0001: fix GIT_* environment variable check under --valgrind

When a test case is run without --valgrind, the wrap-for-bin.sh
helper script inserts the environment variable GIT_TEXTDOMAINDIR, but
when run with --valgrind, the variable is missing. A recently
introduced test case expects the presence of the variable, though, and
fails under --valgrind.

Yep.

It's interesting though that valgrind sets up some variables without
going through bin-wrappers. That's understandable because valgrind
support is added (in 4e1be63) 10 months before bin-wrappers (in
ea92519).  But it's probably better that we inject valgrind command
from inside bin-wrappers script, the same way we inject gdb, I think.

I had the same thought and even started on a patch, but it doesn't quite
work. The bin-wrappers are all about intercepting what goes into the
user's $PATH, and pointing our libexec dir at the main build.

So we have "git" and "git-upload-pack" in bin-wrappers, but not
"git-log". Whereas the valgrind code wants to intercept _all_ of the
test script's invocations of git, including ones spawned by scripts,
other git commands, etc. So conceptually, it wants to intercept
$GIT_EXEC_PATH.

-Peff

Am 03.03.2016 um 13:09 schrieb Duy Nguyen:

+the-other-Johannes who added valgrind support.

On Thu, Mar 3, 2016 at 1:55 PM, Johannes Sixt [off-list ref] wrote:

quoted

---- 8< ----
Subject: [PATCH] t0001: fix GIT_* environment variable check under --valgrind

When a test case is run without --valgrind, the wrap-for-bin.sh
helper script inserts the environment variable GIT_TEXTDOMAINDIR, but
when run with --valgrind, the variable is missing. A recently
introduced test case expects the presence of the variable, though, and
fails under --valgrind.

Yep.

It's interesting though that valgrind sets up some variables without
going through bin-wrappers. That's understandable because valgrind
support is added (in 4e1be63) 10 months before bin-wrappers (in
ea92519).  But it's probably better that we inject valgrind command
from inside bin-wrappers script, the same way we inject gdb, I think.

quoted

Rewrite the test case to strip conditially defined environment variables
from both expected and actual output.

Or we could set GIT_TEXTDOMAINDIR in the "if test -n $valgrind" code
in test-lib.sh, which makes the two more consistent. Also simpler
patch.

My fix (or something along its lines) is needed nevertheless. Just 
s/--valgrind/--with-dashes/g in the commit message if you want to fix 
the --valgrind case differently ;-)

I run tests on Windows --with-dashes in the hopes that it saves a fork 
and exec or two on every git invocation.

-- Hannes