[PATCH] git-prompt.sh: Don't error on null ${ZSH,BASH}_VERSION

[PATCH] git-prompt.sh: Don't error on null ${ZSH,BASH}_VERSION

[Ville Skyttä]

When the shell is in "nounset" or "set -u" mode, referencing unset or
null variables results in an error. Protect $ZSH_VERSION and
$BASH_VERSION against that.

Signed-off-by: Ville Skyttä <ville.skytta@iki.fi>
---
 contrib/completion/git-prompt.sh | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/contrib/completion/git-prompt.sh b/contrib/completion/git-prompt.sh
index 64219e6..341cf6b 100644
--- a/contrib/completion/git-prompt.sh
+++ b/contrib/completion/git-prompt.sh
@@ -355,8 +355,8 @@ __git_ps1 ()
 	# incorrect.)
 	#
 	local ps1_expanded=yes
-	[ -z "$ZSH_VERSION" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
-	[ -z "$BASH_VERSION" ] || shopt -q promptvars || ps1_expanded=no
+	[ -z "${ZSH_VERSION-}" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
+	[ -z "${BASH_VERSION-}" ] || shopt -q promptvars || ps1_expanded=no
 
 	local repo_info rev_parse_exit_code
 	repo_info="$(git rev-parse --git-dir --is-inside-git-dir \
-- 
2.5.5

Re: [PATCH] git-prompt.sh: Don't error on null ${ZSH,BASH}_VERSION

[Junio C Hamano]

Ville Skyttä <ville.skytta@iki.fi> writes:

> When the shell is in "nounset" or "set -u" mode, referencing unset or
> null variables results in an error. Protect $ZSH_VERSION and
> $BASH_VERSION against that.
>
> Signed-off-by: Ville Skyttä <ville.skytta@iki.fi>
> ---

Makes sense from a cursory look.  Thanks.

>  contrib/completion/git-prompt.sh | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/contrib/completion/git-prompt.sh b/contrib/completion/git-prompt.sh
> index 64219e6..341cf6b 100644
> --- a/contrib/completion/git-prompt.sh
> +++ b/contrib/completion/git-prompt.sh
> @@ -355,8 +355,8 @@ __git_ps1 ()
>  	# incorrect.)
>  	#
>  	local ps1_expanded=yes
> -	[ -z "$ZSH_VERSION" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
> -	[ -z "$BASH_VERSION" ] || shopt -q promptvars || ps1_expanded=no
> +	[ -z "${ZSH_VERSION-}" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
> +	[ -z "${BASH_VERSION-}" ] || shopt -q promptvars || ps1_expanded=no
>  
>  	local repo_info rev_parse_exit_code
>  	repo_info="$(git rev-parse --git-dir --is-inside-git-dir \

Re: [PATCH] git-prompt.sh: Don't error on null ${ZSH,BASH}_VERSION

[Junio C Hamano]

Ville Skyttä <ville.skytta@iki.fi> writes:

> When the shell is in "nounset" or "set -u" mode, referencing unset or
> null variables results in an error. Protect $ZSH_VERSION and
> $BASH_VERSION against that.
>  	local ps1_expanded=yes
> -	[ -z "$ZSH_VERSION" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
> -	[ -z "$BASH_VERSION" ] || shopt -q promptvars || ps1_expanded=no
> +	[ -z "${ZSH_VERSION-}" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
> +	[ -z "${BASH_VERSION-}" ] || shopt -q promptvars || ps1_expanded=no

I wonder if we can teach t9903 to run (at least) some parts of the
completion script under "set -u" to catch these automatically without
a report from an end user.

With the attached "just started to scratch the surface" patch to
test-lib.sh and fix by Ville to the git-prompt.sh script, we find
another one.

    ...contrib/completion/git-prompt.sh: line 481: short_sha: unbound variable

Ville, I think you want to take the "git-prompt.sh" part from the
following and replace it with your patch.

Others, I am not sure how seriously we would want to support those
who use "set -u"; the damage to test-lib.sh so far (see below) does
not look _too_ bad, but the reason why I have "set -u" after
including lib-bash.sh is because it will trigger a lot more errors
while lib-bash and test-lib do their preparation.

 contrib/completion/git-prompt.sh | 5 +++--
 t/t9903-bash-prompt.sh           | 2 ++
 t/test-lib.sh                    | 8 +++++++-
 3 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/contrib/completion/git-prompt.sh b/contrib/completion/git-prompt.sh
index 64219e6..ed994a9 100644
--- a/contrib/completion/git-prompt.sh
+++ b/contrib/completion/git-prompt.sh
@@ -355,8 +355,8 @@ __git_ps1 ()
 	# incorrect.)
 	#
 	local ps1_expanded=yes
-	[ -z "$ZSH_VERSION" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
-	[ -z "$BASH_VERSION" ] || shopt -q promptvars || ps1_expanded=no
+	[ -z "${ZSH_VERSION-}" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
+	[ -z "${BASH_VERSION-}" ] || shopt -q promptvars || ps1_expanded=no
 
 	local repo_info rev_parse_exit_code
 	repo_info="$(git rev-parse --git-dir --is-inside-git-dir \
@@ -369,6 +369,7 @@ __git_ps1 ()
 	fi
 
 	local short_sha
+	short_sha=
 	if [ "$rev_parse_exit_code" = "0" ]; then
 		short_sha="${repo_info##*$'\n'}"
 		repo_info="${repo_info%$'\n'*}"
diff --git a/t/t9903-bash-prompt.sh b/t/t9903-bash-prompt.sh
index 0db4469..96b5ff8 100755
--- a/t/t9903-bash-prompt.sh
+++ b/t/t9903-bash-prompt.sh
@@ -7,6 +7,8 @@ test_description='test git-specific bash prompt functions'
 
 . ./lib-bash.sh
 
+set -u
+
 . "$GIT_BUILD_DIR/contrib/completion/git-prompt.sh"
 
 actual="$TRASH_DIRECTORY/actual"
diff --git a/t/test-lib.sh b/t/test-lib.sh
index 0055ebb..36ca13e 100644
--- a/t/test-lib.sh
+++ b/t/test-lib.sh
@@ -186,6 +186,9 @@ test "x$TERM" != "xdumb" && (
 	) &&
 	color=t
 
+HARNESS_ACTIVE= quiet= verbose_only= run_list= trace=
+: ${GIT_SKIP_TESTS=}
+
 while test "$#" -ne 0
 do
 	case "$1" in
@@ -585,7 +588,7 @@ test_eval_ () {
 
 test_run_ () {
 	test_cleanup=:
-	expecting_failure=$2
+	expecting_failure=${2-}
 
 	if test "${GIT_TEST_CHAIN_LINT:-1}" != 0; then
 		# turn off tracing for this test-eval, as it simply creates
@@ -750,6 +753,8 @@ test_done () {
 	esac
 }
 
+GIT_VALGRIND=
+
 if test -n "$valgrind"
 then
 	make_symlink () {
@@ -902,6 +907,7 @@ fi
 # in subprocesses like git equals our $PWD (for pathname comparisons).
 cd -P "$TRASH_DIRECTORY" || exit 1
 
+skip_all=
 this_test=${0##*/}
 this_test=${this_test%%-*}
 if match_pattern_list "$this_test" $GIT_SKIP_TESTS

Re: [PATCH] git-prompt.sh: Don't error on null ${ZSH,BASH}_VERSION

[Junio C Hamano]

Junio C Hamano <gitster@pobox.com> writes:

> Others, I am not sure how seriously we would want to support those
> who use "set -u"; the damage to test-lib.sh so far (see below) does
> not look _too_ bad, but the reason why I have "set -u" after
> including lib-bash.sh is because it will trigger a lot more errors
> while lib-bash and test-lib do their preparation.

Ehh, this didn't come out very well as I hoped, primarily because my
thinking behind it was muddy X-<.

I do want to see us keep supporting those who "set -u" in their
interactive shell environment, and we should strive to make sure
that the "prompt" and the "completion", which are part of their
interactive shell experience, work well with "set -u".

It does not mean "cd t && sh t$anything.sh" must be "set -u" clean
at all (this is where my thinking was muddy); it is sufficient if we
can successfully run t9903 (prompt) with "set -u" before the script
includes "git-prompt".  We do not have to be able to insert "set -u"
at the beginning of test-lib.sh and cleanly run other tests.

Having said that, some parts of "minimum" change shown in my "how
about this" patch would probably be worth doing anyway.  They make
sure that:

 (1) variables the test framework uses will always be initialized
     properly, without getting affected a stray environment variable
     the user may have.  E.g. "trace".

 (2) environment variables the test framework may accept to change
     the behaviour, e.g. GIT_SKIP_TESTS, are clearly documented by
     using the idiom "set to an empty string if not set".

So in that sense, even though we may not add "set -u" at the
beginning of test-lib.sh in the production code, doing so may be a
good developer aid to catch other variables the attached patch did
not catch and complete the above two points.

> diff --git a/t/test-lib.sh b/t/test-lib.sh
> index 0055ebb..36ca13e 100644
> --- a/t/test-lib.sh
> +++ b/t/test-lib.sh
> @@ -186,6 +186,9 @@ test "x$TERM" != "xdumb" && (
>  	) &&
>  	color=t
>  
> +HARNESS_ACTIVE= quiet= verbose_only= run_list= trace=
> +: ${GIT_SKIP_TESTS=}
> +
>  while test "$#" -ne 0
>  do
>  	case "$1" in
> @@ -585,7 +588,7 @@ test_eval_ () {
>  
>  test_run_ () {
>  	test_cleanup=:
> -	expecting_failure=$2
> +	expecting_failure=${2-}
>  
>  	if test "${GIT_TEST_CHAIN_LINT:-1}" != 0; then
>  		# turn off tracing for this test-eval, as it simply creates
> @@ -750,6 +753,8 @@ test_done () {
>  	esac
>  }
>  
> +GIT_VALGRIND=
> +
>  if test -n "$valgrind"
>  then
>  	make_symlink () {
> @@ -902,6 +907,7 @@ fi
>  # in subprocesses like git equals our $PWD (for pathname comparisons).
>  cd -P "$TRASH_DIRECTORY" || exit 1
>  
> +skip_all=
>  this_test=${0##*/}
>  this_test=${this_test%%-*}
>  if match_pattern_list "$this_test" $GIT_SKIP_TESTS

Re: [PATCH] git-prompt.sh: Don't error on null ${ZSH,BASH}_VERSION

[Ville Skyttä]

On Tue, May 24, 2016 at 8:27 PM, Junio C Hamano <gitster@pobox.com> wrote:
> Ville Skyttä <ville.skytta@iki.fi> writes:
>
>> When the shell is in "nounset" or "set -u" mode, referencing unset or
>> null variables results in an error. Protect $ZSH_VERSION and
>> $BASH_VERSION against that.
>>       local ps1_expanded=yes
>> -     [ -z "$ZSH_VERSION" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
>> -     [ -z "$BASH_VERSION" ] || shopt -q promptvars || ps1_expanded=no
>> +     [ -z "${ZSH_VERSION-}" ] || [[ -o PROMPT_SUBST ]] || ps1_expanded=no
>> +     [ -z "${BASH_VERSION-}" ] || shopt -q promptvars || ps1_expanded=no
>
> I wonder if we can teach t9903 to run (at least) some parts of the
> completion script under "set -u" to catch these automatically without
> a report from an end user.
>
> With the attached "just started to scratch the surface" patch to
> test-lib.sh and fix by Ville to the git-prompt.sh script, we find
> another one.
>
>     ...contrib/completion/git-prompt.sh: line 481: short_sha: unbound variable
>
> Ville, I think you want to take the "git-prompt.sh" part from the
> following and replace it with your patch.

Hm, I'm not entirely sure what I was expected to do, but I took the
short_sha bit, added it to my patch (modifying to be in line with the
style of the rest of the file), and submitted a superseding patch with
the subject "[PATCH 1/2] git-prompt.sh: Don't error on null
${ZSH,BASH}_VERSION, $short_sha" (and there's no part 2/2 coming, that
was a user error).