Closed p5pRT closed 8 years ago
Ive got patches against blead to collect more timing data using TEST_HARNESS;
measures and prints elapsed time for each test (current functionality) HARNESS_TIMER=1 make test; t/io/nargv ................ ok 18 ms
calls times() before and after each test\, prints child user\, system times HARNESS_TIMER=2 make test; t/comp/colon .............. ok 17 ms 10 ms 0 ms t/comp/decl ............... ok 4 ms 0 ms 0 ms t/comp/final_line_num ..... ok 5 ms 0 ms 10 ms t/comp/fold ............... ok 10 ms 10 ms 0 ms t/comp/form_scope ......... ok 8 ms 0 ms 10 ms
calls times() at END of each test\, passes 4 measures back to t/TEST as diag-output t/TEST prints them at end of line for each tested file. HARNESS_TIMER=3 make test; also writes all collected timing data to Storable file in the directory. HARNESS_TIMER=../../perf make test; # (existing directory) t/cmd/elsif ............... ok 9 ms 0 ms 0 ms # ET: (ms) u:0 s:0 cu:0 cs:0 t/cmd/for ................. ok 19 ms 20 ms 0 ms # ET: (ms) u:10 s:0 cu:0 cs:0 t/cmd/mod ................. ok 5 ms 0 ms 10 ms # ET: (ms) u:0 s:0 cu:0 cs:0 t/cmd/subval .............. ok 14 ms 0 ms 0 ms # ET: (ms) u:0 s:0 cu:0 cs:0
patches attached.
On Wed Sep 07 16:54:48 2011\, yoduh wrote:
patches attached.
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I have reservations about the fifth patch. I thought that TestInit.pm was supposed to be written in such a way that most test scripts don’t need to take it into account. But if it has an END block\, things start to get complicated. Test authors will now have to worry about whether they need POSIX::_exit\, whether that’s portable\, etc.
The RT System itself - Status changed from 'new' to 'open'
On Thu\, Sep 8\, 2011 at 3:40 PM\, Father Chrysostomos via RT \perlbug\-followup@​perl\.org wrote:
On Wed Sep 07 16:54:48 2011\, yoduh wrote:
patches attached.
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I have reservations about the fifth patch. I thought that TestInit.pm was supposed to be written in such a way that most test scripts don’t need to take it into account. But if it has an END block\, things start to get complicated. Test authors will now have to worry about whether they need POSIX::_exit\, whether that’s portable\, etc.
it gave me a little heartburn too\, esp when I had to add untie STDOUT to get a few tests to pass. But the END block is a noop unless HARNESS_TIMER is set\, and as comment notes\, the condition could be tighter still.
Ive no objection to it sitting in a smoke-me branch for a while\, but in that case\, HARNESS_TIMER should be forced to a value that activates the END block.
The big knock against perlbench is the variance of the results\, systematic performance differences are lost in the noise. Having lots of detailed data per run\, esp if collected by default\, could provide the raw data for statistical methods to reduce the uncertainty inherent in the noise. This doesnt defend patch 5 per se (Storable file is in patch 4)\, but the additional data does get into multi-process tests\, and should also work better in parallel tests.
FWIW\, Ive been running this for a while with TEST_HARNESS=../../perf\, though I dont always look at the results (goal was to pool lots of timing data).
[jimc@groucho perl]$ ls ../perf |wc 569 569 15643
If the patchset gets traction\, I'll look to add a bit more data to it: - rc & fails to each test-file record - test-command; make test vs make test.valgrind produce dramatically different numbers and theres nothing in the Storable file to distinguish them. - track subprocesses more closely (this requires patch 5) add ($$\, $0\, @ARGV) to '# ET (ms) ..." message use that to parse and store more data timingdetail => { $$ => { cmd => [$0\, @ARGV]\, u => $usr\, s => $sys\, cu=> $childusr\, cs => $childsys }}
thanks ~jimc
On Fri Sep 09 10:41:35 2011\, yoduh wrote:
On Thu\, Sep 8\, 2011 at 3:40 PM\, Father Chrysostomos via RT \perlbug\-followup@​perl\.org wrote:
On Wed Sep 07 16:54:48 2011\, yoduh wrote:
patches attached.
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I have reservations about the fifth patch. I thought that TestInit.pm was supposed to be written in such a way that most test scripts don’t need to take it into account. But if it has an END block\, things start to get complicated. Test authors will now have to worry about whether they need POSIX::_exit\, whether that’s portable\, etc.
it gave me a little heartburn too\, esp when I had to add untie STDOUT to get a few tests to pass. But the END block is a noop unless HARNESS_TIMER is set\, and as comment notes\, the condition could be tighter still.
Ive no objection to it sitting in a smoke-me branch for a while\, but in that case\, HARNESS_TIMER should be forced to a value that activates the END block.
I’ve applied the first four patches. Thank you. Their commit IDs are:
f7b9b043ef957755618188c0c17f55142f7a3506 b49055e966b614a7aa5d3eda7075ca3262694b6f 8e03ad8f6c24ab6b98b196deca21ee507e4773f7 25a2b27fba4a91d93a7517d10647a68981f2453f
The fifth patch I’m going to leave for someone who understands the test infrastructure a lot better. (Nick\, if you are reading this\, could you please have a look? :-)
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
On Wed Sep 07 16:54:48 2011\, yoduh wrote:
patches attached.
Gah. I don't like the interaction of Jesse's desire for patches to be sent to RT\, with the usual solution of attaching patches *to* RT. It makes them a pain to review because
a: they are now (at least) a click away. (or more\, if my web browser insists that I should download them and fire up a text editor to look at them) (Or infinitely far away if I have an offline mailer\, and I'm on a $DHH plane) b: it's not trivial to comment on them in an e-mail message\, which is the workflow round here. Or at least\, has been
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
VALGRIND ?= valgrind VG_TEST ?= ./perl -e 1 2>/dev/null
in Makefile\, as it is (I think) a GNU make-ism. But nothing seems to have choked on it so far.
[I think valgrind only works well on platforms that have gmake as standard (OS X and Linux)\, so possibly Makefile.SH should write this out conditionally based on which make we have.]
I have reservations about the fifth patch. I thought that TestInit.pm was supposed to be written in such a way that most test scripts don't need to take it into account. But if it has an END block\, things start to get complicated. Test authors will now have to worry about whether they need POSIX::_exit\, whether that's portable\, etc.
That's a very good point. The way it's written\, it looks to be sane\, safe and clean\, but if we keep adding functionality to a module that's supposed to "not be there"\, we keep increasing the risk of some side effect somehow getting something "wrong"\, tearing the invisibility cloak.
Given that all this code is controlled by t/TEST\, which is also controlling the command line passed to the test\, I think it would equally possible to implement the END injection in its own module. That way it would only added to the test's command line with -M by t/TEST\, conclusively isolating it as a cause of any problems in the general case. That also means that TestInit continues only to be about "Init". Which I think I find pleasing.
Nicholas Clark
On Fri\, Sep 09\, 2011 at 11:40:30AM -0600\, Jim Cromie wrote:
The big knock against perlbench is the variance of the results\, systematic performance differences are lost in the noise. Having lots of detailed data per run\, esp if collected by default\, could provide the raw data for statistical methods to reduce the uncertainty inherent in the noise. This doesnt defend patch 5
Doesn't that reasoning for "lots of data and then stats" also apply to perlbench - run it many times and then use statistical methods?
Whilst it's potentially useful to know if the regression tests start taking significantly different time\, I'm still not convinced that they make a good benchmark suite. They serve different purposes:
regression tests
* try to test obscure corner cases * focus on one thing in isolation * should run as quickly as possible\, to avoid programmers getting bored or complacent * often end up being dominated by startup time
benchmarks should
* focus on common code * perform complex behaviour using multiple features * should stress things with the scale of data needed to detect real problems [such as a change to O(bad) behaviour where previously it was O(acceptable)] * unless benchmarking startup time\, strive to avoid it influencing the result
and I don't think it's useful to try to make the *regression* tests pretend to be a benchmark.
I've not looked at it yet\, but I'm hoping that Steffen Schwingon's work on Benchmark::Perl::Formance is going to produce a more comprehensive benchmark than perlbench:
https://metacpan.org/module/Benchmark::Perl::Formance
Nicholas Clark
On Fri\, Sep 09\, 2011 at 09:42:08PM -0700\, Father Chrysostomos via RT wrote:
infrastructure a lot better. (Nick\, if you are reading this\, could you please have a look? :-)
I am now. I had a backlog. I feel troubled that the backlog grows faster than I can deal with it\, unless I stop writing code and just deal with e-mail full time\, answering questions.
Nicholas Clark
On Wed\, Sep 14\, 2011 at 04:56:27PM +0100\, Nicholas Clark wrote:
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
Actually\, I do have an opinion on the fourth.
Why are we adding this to t/TEST\, which is *supposed* to be the fail-safe way to run the test suite\, and keep working even when sophisticated stuff is broken.
Surely it should be in t/harness? And doesn't some of it start to duplicate what TAP::Harness already does\, such as recording test timings in t/test_state
I don't like the idea of adding functionality to something core only\, when it feels like such functionality could be in a CPAN module\, and hence available to all.
Nicholas Clark
* Nicholas Clark \nick@​ccl4\.org [2011-09-14 18:00]:
a: they are now (at least) a click away. (or more\, if my web browser insists that I should download them and fire up a text editor to look at them)
You can reduce it to just a second click\, at least: https://addons.mozilla.org/en-US/firefox/addon/open-in-browser/ http://spasche.net/openinbrowser/
On Wed\, Sep 14\, 2011 at 06:39:51PM +0200\, Aristotle Pagaltzis wrote:
* Nicholas Clark \nick@​ccl4\.org [2011-09-14 18:00]:
a: they are now (at least) a click away. (or more\, if my web browser insists that I should download them and fire up a text editor to look at them)
You can reduce it to just a second click\, at least: https://addons.mozilla.org/en-US/firefox/addon/open-in-browser/ http://spasche.net/openinbrowser/
I have some addon in Firefox to do this already. I don't know how to force it in Chrome\, which was what I initially used to look at RT.
[And in the general case this would be more useful\, as Chrome has a built-in PDF viewer. Firefox only does text\, HTML and images]
Nicholas Clark
On Wed Sep 14 09:27:09 2011\, nicholas wrote:
On Wed\, Sep 14\, 2011 at 04:56:27PM +0100\, Nicholas Clark wrote:
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
Actually\, I do have an opinion on the fourth.
Why are we adding this to t/TEST\, which is *supposed* to be the fail- safe way to run the test suite\, and keep working even when sophisticated stuff is broken.
Surely it should be in t/harness? And doesn't some of it start to duplicate what TAP::Harness already does\, such as recording test timings in t/test_state
I don't like the idea of adding functionality to something core only\, when it feels like such functionality could be in a CPAN module\, and hence available to all.
Nicholas Clark
Since this is not really an area of perl I feel comfortable about making decisions on\, have I leave the rest of this ticket to you? (There are other outstanding patches.)
On Wed Sep 14 13:32:55 2011\, sprout wrote:
On Wed Sep 14 09:27:09 2011\, nicholas wrote:
Nicholas Clark
Since this is not really an area of perl I feel comfortable about making decisions on\, have I leave the rest of this ticket to you? (There are other outstanding patches.)
s/have I leave/may I leave/
I have reservations about the fifth patch. I thought that TestInit.pm was supposed to be written in such a way that most test scripts don't need to take it into account. But if it has an END block\, things start to get complicated. Test authors will now have to worry about whether they need POSIX::_exit\, whether that's portable\, etc.
That's a very good point. The way it's written\, it looks to be sane\, safe and clean\, but if we keep adding functionality to a module that's supposed to "not be there"\, we keep increasing the risk of some side effect somehow getting something "wrong"\, tearing the invisibility cloak.
Given that all this code is controlled by t/TEST\, which is also controlling the command line passed to the test\, I think it would equally possible to implement the END injection in its own module. That way it would only added to the test's command line with -M by t/TEST\, conclusively isolating it as a cause of any problems in the general case. That also means that TestInit continues only to be about "Init". Which I think I find pleasing.
Nicholas Clark
OK. on a 2nd reading\, I think the light went on.
under some modes t/TEST will issue each subtest with $tperl -MTimingHarness:@args $file_to_test
possibly triggered by HARNESS_TIMER="-MTimingHarness:@args"
t/TEST will still have to do some work itself\, like gathering config\,platform\,etc\, but that can also be defined in the TimingHarness.pm\, and called from t/TEST with a single statement.
pre_DUT: start unit timing\, post_DUT: collect timing\, etc\, write to Sqllite db-file
test_run_start: starts overall elapsed time\, setup DB test_run_end: gathers config\, platform info and test-state
This looks to be easier to add to t/harness too.
I'll try this out.. thanks.
On Wed\, Sep 14\, 2011 at 9:56 AM\, Nicholas Clark \nick@​ccl4\.org wrote:
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
VALGRIND ?= valgrind VG_TEST ?= ./perl -e 1 2>/dev/null
in Makefile\, as it is (I think) a GNU make-ism. But nothing seems to have choked on it so far.
So. VMS chokes on '?=' (sorry Craig) Anyone know a portable form ?
[I think valgrind only works well on platforms that have gmake as standard (OS X and Linux)\, so possibly Makefile.SH should write this out conditionally based on which make we have.]
I'll do this\, if nobody beats me to it.
Rather than checking platform\, is it enough to check for the valgrind executable ? Its a pretty good proxy for a devbox. Its not perfect for a test.perf target (OS X wont have it)\, but works on linux - and if perf isnt installed\, make test.perf will clearly say so.
On Fri\, Sep 16\, 2011 at 3:01 PM\, Jim Cromie \jim\.cromie@​gmail\.com wrote:
On Wed\, Sep 14\, 2011 at 9:56 AM\, Nicholas Clark \nick@​ccl4\.org wrote:
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
VALGRIND ?= valgrind VG_TEST ?= ./perl -e 1 2>/dev/null
in Makefile\, as it is (I think) a GNU make-ism. But nothing seems to have choked on it so far.
So. VMS chokes on '?=' (sorry Craig) Anyone know a portable form ?
I think he meant make that is not GNU make. VMS does not use Makefile.SH nor is it currently viable to build with any flavor of make. Nor does valgrind exist that I am aware of\, though there is a native heap analyzer.
On Fri\, Sep 16\, 2011 at 2:25 PM\, Craig A. Berry \craig\.a\.berry@​gmail\.com wrote:
On Fri\, Sep 16\, 2011 at 3:01 PM\, Jim Cromie \jim\.cromie@​gmail\.com wrote:
On Wed\, Sep 14\, 2011 at 9:56 AM\, Nicholas Clark \nick@​ccl4\.org wrote:
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
VALGRIND ?= valgrind VG_TEST ?= ./perl -e 1 2>/dev/null
in Makefile\, as it is (I think) a GNU make-ism. But nothing seems to have choked on it so far.
So. VMS chokes on '?=' (sorry Craig) Anyone know a portable form ?
I think he meant make that is not GNU make.
agreed. What form of if-then works in your make ? Maybe its a portable form.
VMS does not use Makefile.SH nor is it currently viable to build with any flavor of make.
So VMS uses pre-packaged Makefile\, and you hand edited out the "?=" ?
Nor does valgrind exist that I am aware of\, though there is a native heap analyzer.
Hmm. Since Makefile.SH is unused on VMS\, it cant skip writing the linux related targets. maybe tailoring Makefile.SH output per platform isnt worth it.
On Fri\, Sep 16\, 2011 at 3:44 PM\, Jim Cromie \jim\.cromie@​gmail\.com wrote:
On Fri\, Sep 16\, 2011 at 2:25 PM\, Craig A. Berry \craig\.a\.berry@​gmail\.com wrote:
On Fri\, Sep 16\, 2011 at 3:01 PM\, Jim Cromie \jim\.cromie@​gmail\.com wrote:
On Wed\, Sep 14\, 2011 at 9:56 AM\, Nicholas Clark \nick@​ccl4\.org wrote:
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
VALGRIND ?= valgrind VG_TEST ?= ./perl -e 1 2>/dev/null
in Makefile\, as it is (I think) a GNU make-ism. But nothing seems to have choked on it so far.
So. VMS chokes on '?=' (sorry Craig) Anyone know a portable form ?
I think he meant make that is not GNU make.
agreed. What form of if-then works in your make ? Maybe its a portable form.
.IF "$(FOO)" .EQ "BAR" BAZ = $(FOO) .ELSE BAZ = .ENDIF
but that's unlikely to be relevant to what you're interested in doing.
VMS does not use Makefile.SH nor is it currently viable to build with any flavor of make.
So VMS uses pre-packaged Makefile\, and you hand edited out the "?=" ?
The VMS port uses an independent configuration and semi-independent build system. The key files are configure.com and vms/descrip_mms.template\, the latter of which gets processed by the former into a description file called descrip.mms\, which is very similar to a Makefile but is not exactly the same thing. It is then processed by HP's MMS (Module Management System) or the freeware equivalent MMK\, which are make-like tools but are not make.
That's more than you wanted to know\, but the point is you can't hurt the VMS build by mangling Makefile.SH. I think the portability concern Nicholas raised has entirely to do with various unixen that have their own\, non-GNU make.
On Wed\, Sep 14\, 2011 at 01:32:56PM -0700\, Father Chrysostomos via RT wrote:
Since this is not really an area of perl I feel comfortable about making decisions on\, have I leave the rest of this ticket to you? (There are other outstanding patches.)
I thought that only the fifth patch was outstanding\, and I'd commented on it\, that there was probably a better approach.
Nicholas Clark
On Fri\, Sep 16\, 2011 at 04:27:03PM -0500\, Craig A. Berry wrote:
On Fri\, Sep 16\, 2011 at 3:44 PM\, Jim Cromie \jim\.cromie@​gmail\.com wrote:
On Fri\, Sep 16\, 2011 at 2:25 PM\, Craig A. Berry \craig\.a\.berry@​gmail\.com wrote:
On Fri\, Sep 16\, 2011 at 3:01 PM\, Jim Cromie \jim\.cromie@​gmail\.com wrote:
On Wed\, Sep 14\, 2011 at 9:56 AM\, Nicholas Clark \nick@​ccl4\.org wrote:
On Thu\, Sep 08\, 2011 at 02:40:12PM -0700\, Father Chrysostomos via RT wrote:
I see nothing wrong with the first four patches. Does anyone else have an opinion?
I'm troubled by the unconditional use of
VALGRIND ?= valgrind VG_TEST ?= ./perl -e 1 2>/dev/null
in Makefile\, as it is (I think) a GNU make-ism. But nothing seems to have choked on it so far.
So. VMS chokes on '?=' (sorry Craig) Anyone know a portable form ?
I think he meant make that is not GNU make.
agreed. What form of if-then works in your make ? Maybe its a portable form.
.IF "$(FOO)" .EQ "BAR" BAZ = $(FOO) .ELSE BAZ = .ENDIF
but that's unlikely to be relevant to what you're interested in doing.
I don't think that there *is* a portable form.
That's more than you wanted to know\, but the point is you can't hurt the VMS build by mangling Makefile.SH. I think the portability concern Nicholas raised has entirely to do with various unixen that have their own\, non-GNU make.
ie most of them. :-)
I think that only Linux\, OS X (and I guess Hurd\, if anyone uses that)) have a "make" which is actually GNU make. It seems that the *BSD makes (at least current-ish FreeBSD\, NetBSD and OpenBSD) have ?=
Not sure how many OSes have it as an option in some way\, but we can't rely on it being there.
Which seems to boil down to 3 solutions:
1: Change Makefile.SH Write a Makefile with ?= on Linux and OS X\, and with = everywhere else. (Which really doesn't make any difference\, because valgrind only really works on Linux and OS X 2: Just write = everywhere\, and document that this form of override works:
make -j20 test.valgrind VALGRIND=/bin/false
but using environment variables doesn't:
VALGRIND=/bin/false make test.valgrind
3: Redo the duplicates for $(VALGRIND) and $(VG_TEST) everywhere they are used. ie expand an empty string using shell conditionals in the relevant Makefile rules\, and in t/TEST. [actually\, that one isn't *evil* as we already have this in t/TEST: my $valgrind_exe = $ENV{VALGRIND} // 'valgrind';
and it's simpler than I first thought if we rule that one must override VALGRIND if one overrides VG_TEST]
Nicholas Clark
On Tue Sep 20 07:17:08 2011\, nicholas wrote:
On Wed\, Sep 14\, 2011 at 01:32:56PM -0700\, Father Chrysostomos via RT wrote:
Since this is not really an area of perl I feel comfortable about making decisions on\, have I leave the rest of this ticket to you? (There are other outstanding patches.)
I thought that only the fifth patch was outstanding\, and I'd commented on it\, that there was probably a better approach.
But you had some qualms about the fourth. I thought you might want to modify or revert it.
--
Father Chrysostomos
On Sat Nov 26 19:42:52 2011\, sprout wrote:
On Tue Sep 20 07:17:08 2011\, nicholas wrote:
On Wed\, Sep 14\, 2011 at 01:32:56PM -0700\, Father Chrysostomos via RT wrote:
Since this is not really an area of perl I feel comfortable about making decisions on\, have I leave the rest of this ticket to you? (There are other outstanding patches.)
I thought that only the fifth patch was outstanding\, and I'd commented on it\, that there was probably a better approach.
But you had some qualms about the fourth. I thought you might want to modify or revert it.
There has been no discussion in this ticket in nearly three years.
If I read the ticket correctly\, 4 of the OP's original patches were applied; the fifth was more controversial and was never resolved.
I recommend that we close this ticket and ask anyone who wants to continue the discussion to submit a new RT.
Thank you very much.
-- James E Keenan (jkeenan@cpan.org)
On Sun\, Oct 19\, 2014 at 2:48 AM\, James E Keenan via RT \< perlbug-followup@perl.org> wrote:
There has been no discussion in this ticket in nearly three years.
If I read the ticket correctly\, 4 of the OP's original patches were applied; the fifth was more controversial and was never resolved.
I recommend that we close this ticket and ask anyone who wants to continue the discussion to submit a new RT.
Thank you very much.
Actually\, Jarkko recently supplied a patch to achieve just this\, and very recently released it as Test::Harness 3.34 :-)
Leon
On Sun Nov 09 12:55:00 2014\, LeonT wrote:
On Sun\, Oct 19\, 2014 at 2:48 AM\, James E Keenan via RT \< perlbug-followup@perl.org> wrote:
There has been no discussion in this ticket in nearly three years.
If I read the ticket correctly\, 4 of the OP's original patches were applied; the fifth was more controversial and was never resolved.
I recommend that we close this ticket and ask anyone who wants to continue the discussion to submit a new RT.
Thank you very much.
Actually\, Jarkko recently supplied a patch to achieve just this\, and very recently released it as Test::Harness 3.34 :-)
Leon
Does this mean we can close this ticket now?
On Mon\, Feb 22\, 2016 at 10:46 PM\, l.mai@web.de via RT \< perlbug-followup@perl.org> wrote:
On Sun Nov 09 12:55:00 2014\, LeonT wrote:
On Sun\, Oct 19\, 2014 at 2:48 AM\, James E Keenan via RT \< perlbug-followup@perl.org> wrote:
There has been no discussion in this ticket in nearly three years.
If I read the ticket correctly\, 4 of the OP's original patches were applied; the fifth was more controversial and was never resolved.
I recommend that we close this ticket and ask anyone who wants to continue the discussion to submit a new RT.
Thank you very much.
Actually\, Jarkko recently supplied a patch to achieve just this\, and very recently released it as Test::Harness 3.34 :-)
Leon
Does this mean we can close this ticket now?
--- via perlbug: queue: perl5 status: open https://rt-archive.perl.org/perl5/Ticket/Display.html?id=98662
Yes
Leon
@mauke - Status changed from 'open' to 'resolved'
Migrated from rt.perl.org#98662 (status was 'resolved')
Searchable as RT98662$