Proposed fix for test failures due to long socket paths

Proposed fix for test failures due to long socket paths

[David Bremner]

Amadeusz Żołnowski found a bug in the test suite that causes gpg
failures when the path of the test directory is sufficiently long.
His current solution in Gentoo is to move the sockets into /tmp. It
seems cleaner to enable gnupg's built in /run based short pathed
sockets.  As far as I know the gpgconf option used here is available
in gnupg 2.1.13 and later. Amadeusz reported that the problem was
"fixed" in Gentoo by downgrading gnupg to 2.1.15. So while I'm not
100% sure, it seems like a fix for very recent gnupg is all that is
required. Testing of these patches with older gpg would be
particularly welcome.

[PATCH 1/2] configure: add test for gpgconf --create-socketdir

[David Bremner]

This is primarily intended for use in the test suite (since notmuch
builds fine without gnupg installed). Thus we only write the variable
to sh.config.
---
 configure | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/configure b/configure
index f1773044..46063b5d 100755
--- a/configure
+++ b/configure
@@ -602,6 +602,19 @@ if [ $WITH_DOCS = "1" ] ; then
     fi
 fi
 
+have_gpgconf_socketdir=0
+printf "Checking for gpgconf --{create,remove}-socketdir... "
+if gpgconf --dump-options > /dev/null ; then
+    if gpgconf --dump-options | grep -q create-socketdir ; then
+	printf "Yes.\n"
+	have_gpgconf_socketdir=1
+    else
+	printf "No.\n"
+    fi
+else
+    printf "No. (missing or broken gpgconf?)\n"
+fi
+
 libdir_in_ldconfig=0
 
 printf "Checking which platform we are on... "
@@ -1179,6 +1192,9 @@ NOTMUCH_PYTHON=${python}
 # Are the ruby development files (and ruby) available? If not skip
 # building/testing ruby bindings.
 NOTMUCH_HAVE_RUBY_DEV=${have_ruby_dev}
+
+# Recent enough gnupg to support gpgconf --create-socketdir?
+NOTMUCH_HAVE_GPGCONF_SOCKETDIR=${have_gpgconf_socketdir}
 EOF
 
 # Finally, after everything configured, inform the user how to continue.
-- 
2.11.0

Re: [PATCH 1/2] configure: add test for gpgconf --create-socketdir

[David Bremner]

Note that Amadeusz's actual address is aidecoe@aidecoe.name, you might
want to correct in any followups.

d

[PATCH 2/2] test: use gpgconf --create-socketdir if available

[David Bremner]

This enables the shortened socket pathes in /run or equivalent. The
explicit call to gpgconf is needed for nonstandard GNUPGHOME settings.
---
 test/test-lib.sh | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/test/test-lib.sh b/test/test-lib.sh
index 86e792a8..9aa6163b 100644
--- a/test/test-lib.sh
+++ b/test/test-lib.sh
@@ -229,9 +229,18 @@ test_fixed=0
 test_broken=0
 test_success=0
 
+
+_shutdown_gpg_agent () {
+    if [ ${NOTMUCH_HAVE_GPGCONF_SOCKETDIR} == 1 ]; then
+	gpgconf --kill gpg-agent
+	gpgconf --remove-socketdir
+    fi
+}
+
 _exit_common () {
 	code=$?
 	trap - EXIT
+	_shutdown_gpg_agent
 	set +ex
 	rm -rf "$TEST_TMPDIR"
 }
@@ -1276,6 +1285,11 @@ test_init_ () {
 
 . ./test-lib-common.sh || exit 1
 
+# we need the setting of GNUPGHOME in test-lib-common.sh
+if [ ${NOTMUCH_HAVE_GPGCONF_SOCKETDIR} == 1 ]; then
+    gpgconf --create-socketdir
+fi
+
 emacs_generate_script
 
 
-- 
2.11.0

Re: [PATCH 2/2] test: use gpgconf --create-socketdir if available

[Daniel Kahn Gillmor]

[-- Attachment #1: Type: text/plain, Size: 809 bytes --]

On Tue 2017-02-14 16:42:39 -0500, David Bremner wrote:
> This enables the shortened socket pathes in /run or equivalent. The
> explicit call to gpgconf is needed for nonstandard GNUPGHOME settings.

For the current GnuPG 2.1.18, this pair of patches looks reasonable to
me, and it should indeed solve builds in deeply-nested paths.

That said, i currently consider this a workaround to a bug that should
really be fixed in GnuPG itself, as i noted on gnupg-users:

  https://lists.gnupg.org/pipermail/gnupg-users/2017-February/057692.html
  id:87k28rl6o8.fsf@alice.fifthhorseman.net

I'd love to not see this kind of workaround cruft pile up, because it
tends to cause problems in the future.  is there a way that we can mark
this sort of thing for fairly prompt removal once it's fixed upstream?

     --dkg

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]

Re: [PATCH 2/2] test: use gpgconf --create-socketdir if available

[David Bremner]

Daniel Kahn Gillmor <dkg@fifthhorseman.net> writes:

> I'd love to not see this kind of workaround cruft pile up, because it
> tends to cause problems in the future.  is there a way that we can mark
> this sort of thing for fairly prompt removal once it's fixed upstream?

Maybe once a gnupg bug is filed, write to this list with the suggestion
that we back out the workaround when the bug is fixed upstream
(although we have to think about how long to support gnupg
2.1.18. Thanks debian!). We can then tag that message as a bug in
nmbug.

d

Re: [PATCH 2/2] test: use gpgconf --create-socketdir if available

[Tomi Ollila]

On Tue, Feb 14 2017, David Bremner <david@tethera.net> wrote:

> This enables the shortened socket pathes in /run or equivalent. The
> explicit call to gpgconf is needed for nonstandard GNUPGHOME settings.

2½ comments inline:

__> ---
>  test/test-lib.sh | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
>
> diff --git a/test/test-lib.sh b/test/test-lib.sh
> index 86e792a8..9aa6163b 100644
> --- a/test/test-lib.sh
> +++ b/test/test-lib.sh
> @@ -229,9 +229,18 @@ test_fixed=0
>  test_broken=0
>  test_success=0
>  
> +
> +_shutdown_gpg_agent () {
> +    if [ ${NOTMUCH_HAVE_GPGCONF_SOCKETDIR} == 1 ]; then

 '==' is bashishm (well, maybe works in zsh too), standard
with [ ... ] and (builtin) test(1) is just single '=' for equality.

> +	gpgconf --kill gpg-agent
> +	gpgconf --remove-socketdir
> +    fi
> +}
> +
>  _exit_common () {
>  	code=$?
>  	trap - EXIT
> +	_shutdown_gpg_agent

better put `set +ex` _before_ `_shutdown_gpg_agent` so that possible failure
there may not stop exit processing

>  	set +ex
>  	rm -rf "$TEST_TMPDIR"
>  }
> @@ -1276,6 +1285,11 @@ test_init_ () {
>  
>  . ./test-lib-common.sh || exit 1
>  
> +# we need the setting of GNUPGHOME in test-lib-common.sh
> +if [ ${NOTMUCH_HAVE_GPGCONF_SOCKETDIR} == 1 ]; then

== -> = 

> +    gpgconf --create-socketdir
> +fi
> +
>  emacs_generate_script
>  
>  
> -- 
> 2.11.0
>
> _______________________________________________
> notmuch mailing list
> notmuch@notmuchmail.org
> https://notmuchmail.org/mailman/listinfo/notmuch

Re: Proposed fix for test failures due to long socket paths

[David Bremner]

David Bremner <david@tethera.net> writes:

> Amadeusz Żołnowski found a bug in the test suite that causes gpg
> failures when the path of the test directory is sufficiently long.
> His current solution in Gentoo is to move the sockets into /tmp. It
> seems cleaner to enable gnupg's built in /run based short pathed
> sockets.  As far as I know the gpgconf option used here is available
> in gnupg 2.1.13 and later. Amadeusz reported that the problem was
> "fixed" in Gentoo by downgrading gnupg to 2.1.15. So while I'm not
> 100% sure, it seems like a fix for very recent gnupg is all that is
> required. Testing of these patches with older gpg would be
> particularly welcome.

Pushed the series (with Tomi's changes) to release and master

d