diff mbox series

[1/2] tests: don't mess with fd 7 of test helper functions

Message ID 20210221192512.3096291-1-szeder.dev@gmail.com (mailing list archive)
State New, archived
Headers show
Series [1/2] tests: don't mess with fd 7 of test helper functions | expand

Commit Message

SZEDER Gábor Feb. 21, 2021, 7:25 p.m. UTC
In test helper functions exercising git commands, e.g.
'test_must_fail', 'test_env' and friends, we can't access the test
script's original standard error, and, consequently, BUG() doesn't
work as expected.

The root of the issue stems from a5bf824f3b (t: prevent '-x' tracing
from interfering with test helpers' stderr, 2018-02-25), where we
started to use a couple of file descriptor duplications and
redirections to separate the standard error of git commands exercised
in test helper functions from the stderr containing the '-x' trace
output of said helper functions.  To achieve that the git command's
stderr is redirected to the test helper function's fd 7, which was
previously duplicated from the helper's stderr.  Alas, fd 7 was not
the right choice for this purpose, because fd 7 is the original
standard error of the test script, and, consequently, we now can't
send error messages from within such test helper functions directly to
the test script's stderr.  Since BUG() does want to send its error
message there it doesn't work as expected in such test helper
functions, because:

  - If the test helper's stderr were redirected to a file (as is often
    the case e.g. with 'test_must_fail'), then the "bug in the test
    script" error message would end up in that file.

  - If the test script is invoked without any of the verbose options,
    then that error message would get lost to /dev/null, leaving no
    clues about why the test script aborted so suddenly.

We don't have any BUG() calls in such test helper functions yet, but
6a67c75948 (test-lib-functions: restrict test_must_fail usage,
2020-07-07) did start to verify the parameters of 'test_must_fail' and
report the error with:

    echo >&7 "test_must_fail: only 'git' is allowed: $*"
    return 1

instead of BUG() that we usually use to bail out in case of bogus
parameters.

Use fd 6 instead of fd 7 for these '-x' tracing related duplications
and redirections.  It is a better choice for this purpose, because fd
6 is the test script's original standard input, and neither these test
helper functions not the git commands exercised by them should ever
read from the test scipt's stdin, see 781f76b158 (test-lib: redirect
stdin of tests, 2011-12-15).  Update the aforementioned error
reporting in 'test_must_fail' to send the error message to fd 6 as
well; the next patch will update it to use BUG() instead.

The only two other functions that want to directly write to the test
script's stderr are 'test_pause' and 'debug', but they are not
affected by this issue, because, being special "interactive" debug
aids, their fd 7 were not redirected in a5bf824f3b.

Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
---
 t/lib-terminal.sh       |  4 ++--
 t/test-lib-functions.sh | 26 +++++++++++++-------------
 2 files changed, 15 insertions(+), 15 deletions(-)

Comments

Jeff King Feb. 21, 2021, 9:50 p.m. UTC | #1
On Sun, Feb 21, 2021 at 08:25:11PM +0100, SZEDER Gábor wrote:

> The root of the issue stems from a5bf824f3b (t: prevent '-x' tracing
> from interfering with test helpers' stderr, 2018-02-25), where we
> started to use a couple of file descriptor duplications and
> redirections to separate the standard error of git commands exercised
> in test helper functions from the stderr containing the '-x' trace
> output of said helper functions.  To achieve that the git command's
> stderr is redirected to the test helper function's fd 7, which was
> previously duplicated from the helper's stderr.  Alas, fd 7 was not
> the right choice for this purpose, because fd 7 is the original
> standard error of the test script, and, consequently, we now can't
> send error messages from within such test helper functions directly to
> the test script's stderr.  Since BUG() does want to send its error
> message there it doesn't work as expected in such test helper
> functions, because:
> 
>   - If the test helper's stderr were redirected to a file (as is often
>     the case e.g. with 'test_must_fail'), then the "bug in the test
>     script" error message would end up in that file.
> 
>   - If the test script is invoked without any of the verbose options,
>     then that error message would get lost to /dev/null, leaving no
>     clues about why the test script aborted so suddenly.

Makes sense. Well explained.

> Use fd 6 instead of fd 7 for these '-x' tracing related duplications
> and redirections.  It is a better choice for this purpose, because fd
> 6 is the test script's original standard input, and neither these test
> helper functions not the git commands exercised by them should ever
> read from the test scipt's stdin, see 781f76b158 (test-lib: redirect
> stdin of tests, 2011-12-15).  Update the aforementioned error
> reporting in 'test_must_fail' to send the error message to fd 6 as
> well; the next patch will update it to use BUG() instead.

s/scipt/script/ in the paragraph above.

I agree that 6 is probably reasonable. I wonder if it is worth having a
master comment describing the function of various descriptors within the
test suite, so that people know which ones are available for which
purposes.  It is getting awfully crowded in that space. Sadly, I don't
think we can portably use numbers higher than 9 (bash is happy to, but
even dash cannot).

Of course people would have to know to look for said comment, which they
may not do. :)

-Peff
Junio C Hamano Feb. 22, 2021, 5:45 p.m. UTC | #2
Jeff King <peff@peff.net> writes:

> I agree that 6 is probably reasonable. I wonder if it is worth having a
> master comment describing the function of various descriptors within the
> test suite, so that people know which ones are available for which
> purposes.  It is getting awfully crowded in that space.

Thanks for a review.
I had the same impression.  I'll wait for a reroll.
diff mbox series

Patch

diff --git a/t/lib-terminal.sh b/t/lib-terminal.sh
index e3809dcead..454909c087 100644
--- a/t/lib-terminal.sh
+++ b/t/lib-terminal.sh
@@ -9,8 +9,8 @@  test_terminal () {
 		echo >&4 "test_terminal: need to declare TTY prerequisite"
 		return 127
 	fi
-	perl "$TEST_DIRECTORY"/test-terminal.perl "$@" 2>&7
-} 7>&2 2>&4
+	perl "$TEST_DIRECTORY"/test-terminal.perl "$@" 2>&6
+} 6>&2 2>&4
 
 test_lazy_prereq TTY '
 	test_have_prereq PERL &&
diff --git a/t/test-lib-functions.sh b/t/test-lib-functions.sh
index 05dc2cc6be..a40c1c5d83 100644
--- a/t/test-lib-functions.sh
+++ b/t/test-lib-functions.sh
@@ -910,10 +910,10 @@  test_must_fail () {
 	esac
 	if ! test_must_fail_acceptable "$@"
 	then
-		echo >&7 "test_must_fail: only 'git' is allowed: $*"
+		echo >&6 "test_must_fail: only 'git' is allowed: $*"
 		return 1
 	fi
-	"$@" 2>&7
+	"$@" 2>&6
 	exit_code=$?
 	if test $exit_code -eq 0 && ! list_contains "$_test_ok" success
 	then
@@ -936,7 +936,7 @@  test_must_fail () {
 		return 1
 	fi
 	return 0
-} 7>&2 2>&4
+} 6>&2 2>&4
 
 # Similar to test_must_fail, but tolerates success, too.  This is
 # meant to be used in contexts like:
@@ -952,8 +952,8 @@  test_must_fail () {
 # Accepts the same options as test_must_fail.
 
 test_might_fail () {
-	test_must_fail ok=success "$@" 2>&7
-} 7>&2 2>&4
+	test_must_fail ok=success "$@" 2>&6
+} 6>&2 2>&4
 
 # Similar to test_must_fail and test_might_fail, but check that a
 # given command exited with a given exit code. Meant to be used as:
@@ -965,7 +965,7 @@  test_might_fail () {
 test_expect_code () {
 	want_code=$1
 	shift
-	"$@" 2>&7
+	"$@" 2>&6
 	exit_code=$?
 	if test $exit_code = $want_code
 	then
@@ -974,7 +974,7 @@  test_expect_code () {
 
 	echo >&4 "test_expect_code: command exited with $exit_code, we wanted $want_code $*"
 	return 1
-} 7>&2 2>&4
+} 6>&2 2>&4
 
 # test_cmp is a helper function to compare actual and expected output.
 # You can use it like:
@@ -1261,8 +1261,8 @@  test_write_lines () {
 }
 
 perl () {
-	command "$PERL_PATH" "$@" 2>&7
-} 7>&2 2>&4
+	command "$PERL_PATH" "$@" 2>&6
+} 6>&2 2>&4
 
 # Given the name of an environment variable with a bool value, normalize
 # its value to a 0 (true) or 1 (false or empty string) return code.
@@ -1388,13 +1388,13 @@  test_env () {
 				shift
 				;;
 			*)
-				"$@" 2>&7
+				"$@" 2>&6
 				exit
 				;;
 			esac
 		done
 	)
-} 7>&2 2>&4
+} 6>&2 2>&4
 
 # Returns true if the numeric exit code in "$2" represents the expected signal
 # in "$1". Signals should be given numerically.
@@ -1436,9 +1436,9 @@  nongit () {
 		GIT_CEILING_DIRECTORIES=$(pwd) &&
 		export GIT_CEILING_DIRECTORIES &&
 		cd non-repo &&
-		"$@" 2>&7
+		"$@" 2>&6
 	)
-} 7>&2 2>&4
+} 6>&2 2>&4
 
 # convert function arguments or stdin (if not arguments given) to pktline
 # representation. If multiple arguments are given, they are separated by