[RFC PATCH] test-lib: allow storing counts with test harnesses

[RFC PATCH] test-lib: allow storing counts with test harnesses

[Adam Dinwoodie]

Currently, test result files are only stored in test-results/*.counts if
$HARNESS_ACTIVE is not set.  This dates from 8ef1abe550 (test-lib: Don't
write test-results when HARNESS_ACTIVE, 2010-08-11), where the
assumption was that if someone were using a test harness like prove,
that would track results and the count files wouldn't be required.
However, as of 49da404070 (test-lib: show missing prereq summary,
2021-11-20), those files also store the list of git test prerequisites
that were missing during the test run, which isn't something that a
generic test harness like prove can provide.

To allow folk using test harnesses to access the lists of missing
prerequisites, add a --counts argument to test-lib that will keep these
counts files even if a test harness is in use.  This means that a
subsequent call of, say, `make -C t aggregate-results` will report
useful information.

It might be preferable to do make a wider-ranging change, including
storing the missing prerequisites separately from the count files, so
the results can be reported regardless of whether the success/failure
counts are wanted, but that would be more disruptive and more work for
relatively little gain.

Signed-off-by: Adam Dinwoodie <adam@dinwoodie.org>
---

The key reason I'm submitting this as an RFC is that last paragraph:
I've tested the below patch, and it achieves what I'm after (letting me
both use prove and audit missing prerequisites), but it further embeds
the use of the "count" files for recording missing prerequisites, where
it might be preferable to do something a bit more complex that treats
the prerequisite reporting as a separate function, rather than a bolt-on
to the pass/fail/etc. counts.

 t/test-lib.sh | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/t/test-lib.sh b/t/test-lib.sh
index 6db377f68b..bbd9ee0e34 100644
--- a/t/test-lib.sh
+++ b/t/test-lib.sh
@@ -157,6 +157,8 @@ parse_option () {
 	local opt="$1"
 
 	case "$opt" in
+	-c|--c|--co|--cou|--coun|--count|--counts)
+		record_counts=t ;;
 	-d|--d|--de|--deb|--debu|--debug)
 		debug=t ;;
 	-i|--i|--im|--imm|--imme|--immed|--immedi|--immedia|--immediat|--immediate)
@@ -1282,7 +1284,7 @@ test_done () {
 
 	finalize_test_output
 
-	if test -z "$HARNESS_ACTIVE"
+	if test -z "$HARNESS_ACTIVE" || test -n "$record_counts"
 	then
 		mkdir -p "$TEST_RESULTS_DIR"
 
-- 
2.39.0

[PATCH] test-lib: allow storing counts with test harnesses

[Adam Dinwoodie]

Currently, test result files are only stored in test-results/*.counts if
$HARNESS_ACTIVE is not set.  This dates from 8ef1abe550 (test-lib: Don't
write test-results when HARNESS_ACTIVE, 2010-08-11), where the
assumption was that if someone were using a test harness like prove,
that would track results and the count files wouldn't be required.
However, as of 49da404070 (test-lib: show missing prereq summary,
2021-11-20), those files also store the list of git test prerequisites
that were missing during the test run, which isn't something that a
generic test harness like prove can provide.

To allow folk using test harnesses to access the lists of missing
prerequisites, add a --counts argument to test-lib that will keep these
counts files even if a test harness is in use.  This means that a
subsequent call of, say, `make -C t aggregate-results` will report
useful information.

It might be preferable to do make a wider-ranging change, including
storing the missing prerequisites separately from the count files, so
the results can be reported regardless of whether the success/failure
counts are wanted, but that would be more disruptive and more work for
relatively little gain.

Signed-off-by: Adam Dinwoodie <adam@dinwoodie.org>
---

I submitted this as an RFC back in December, and received no comments,
so I'm submitting this as an actual patch now.  My key concern was the
final paragraph above -- embedding using the "count" files for something
other than counts -- but I've mostly convinced myself that refactoring
this code to separate that out is unlikely to actually cause significant
pain.

 t/test-lib.sh | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/t/test-lib.sh b/t/test-lib.sh
index 6db377f68b..bbd9ee0e34 100644
--- a/t/test-lib.sh
+++ b/t/test-lib.sh
@@ -157,6 +157,8 @@ parse_option () {
 	local opt="$1"
 
 	case "$opt" in
+	-c|--c|--co|--cou|--coun|--count|--counts)
+		record_counts=t ;;
 	-d|--d|--de|--deb|--debu|--debug)
 		debug=t ;;
 	-i|--i|--im|--imm|--imme|--immed|--immedi|--immedia|--immediat|--immediate)
@@ -1282,7 +1284,7 @@ test_done () {
 
 	finalize_test_output
 
-	if test -z "$HARNESS_ACTIVE"
+	if test -z "$HARNESS_ACTIVE" || test -n "$record_counts"
 	then
 		mkdir -p "$TEST_RESULTS_DIR"
 

Re: [PATCH] test-lib: allow storing counts with test harnesses

[Jeff King]

On Sat, Mar 04, 2023 at 09:22:20PM +0000, Adam Dinwoodie wrote:

> Currently, test result files are only stored in test-results/*.counts if
> $HARNESS_ACTIVE is not set.  This dates from 8ef1abe550 (test-lib: Don't
> write test-results when HARNESS_ACTIVE, 2010-08-11), where the
> assumption was that if someone were using a test harness like prove,
> that would track results and the count files wouldn't be required.
> However, as of 49da404070 (test-lib: show missing prereq summary,
> 2021-11-20), those files also store the list of git test prerequisites
> that were missing during the test run, which isn't something that a
> generic test harness like prove can provide.
> 
> To allow folk using test harnesses to access the lists of missing
> prerequisites, add a --counts argument to test-lib that will keep these
> counts files even if a test harness is in use.  This means that a
> subsequent call of, say, `make -C t aggregate-results` will report
> useful information.

Your goal seems reasonable. I have to wonder if it is even worth
requiring "--counts" here, though. Even 8ef1abe550 claims that the I/O
from writing the results files is minimal. And certainly I run under
prove with "--root=/some/ram/disk", and I haven't noticed any difference
with and without my usual "--verbose-log", which writes a lot more data
into test-results/.

So would it be worth it to just revert 8ef1abe550, and always store the
meta-files? That's one less option to support, and one less surprise
when some other feature is built around them.

Or is there some reason that we really want to have a mode where nothing
is written into t/? From reading 8ef1abe550 it sounded like this was
mostly a hygiene / optimization thing, and not some special mode we
cared about supporting.

>  t/test-lib.sh | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)

The patch itself looks correct, if we want to go with a --counts option.

-Peff

Re: [PATCH] test-lib: allow storing counts with test harnesses

[Junio C Hamano]

Jeff King <peff@peff.net> writes:

> So would it be worth it to just revert 8ef1abe550, and always store the
> meta-files? That's one less option to support, and one less surprise
> when some other feature is built around them.
>
> Or is there some reason that we really want to have a mode where nothing
> is written into t/? From reading 8ef1abe550 it sounded like this was
> mostly a hygiene / optimization thing, and not some special mode we
> cared about supporting.

Interesting thought.  It would simplify our lives to have fewer
conditionally-moving parts.

>>  t/test-lib.sh | 4 +++-
>>  1 file changed, 3 insertions(+), 1 deletion(-)
>
> The patch itself looks correct, if we want to go with a --counts option.

Yes.

Re: [PATCH] test-lib: allow storing counts with test harnesses

[Elijah Newren]

On Sat, Mar 4, 2023 at 2:16 PM Adam Dinwoodie <adam@dinwoodie.org> wrote:
>
> Currently, test result files are only stored in test-results/*.counts if
> $HARNESS_ACTIVE is not set.  This dates from 8ef1abe550 (test-lib: Don't
> write test-results when HARNESS_ACTIVE, 2010-08-11), where the
> assumption was that if someone were using a test harness like prove,
> that would track results and the count files wouldn't be required.
> However, as of 49da404070 (test-lib: show missing prereq summary,
> 2021-11-20), those files also store the list of git test prerequisites
> that were missing during the test run, which isn't something that a
> generic test harness like prove can provide.
>
> To allow folk using test harnesses to access the lists of missing
> prerequisites, add a --counts argument to test-lib that will keep these
> counts files even if a test harness is in use.  This means that a
> subsequent call of, say, `make -C t aggregate-results` will report
> useful information.
>
> It might be preferable to do make a wider-ranging change, including

Replace "do make" with either "do" or "make"?

> storing the missing prerequisites separately from the count files, so
> the results can be reported regardless of whether the success/failure
> counts are wanted, but that would be more disruptive and more work for
> relatively little gain.
>
> Signed-off-by: Adam Dinwoodie <adam@dinwoodie.org>
> ---
>
> I submitted this as an RFC back in December, and received no comments,
> so I'm submitting this as an actual patch now.  My key concern was the
> final paragraph above -- embedding using the "count" files for something
> other than counts -- but I've mostly convinced myself that refactoring
> this code to separate that out is unlikely to actually cause significant
> pain.

Actual code change looks fine to me as well, though I see Peff has
pointed out we might not even need to make it an option.