Re: Smoke [blead] v5.26.0-1591-g56e48c1076 FAIL(XM) os/390 25.00 (2964/)

[Karl Williamson]

On 12/06/2017 04:21 AM, Yaroslav Kuzmin wrote:
> В Пн., 04/12/2017 в 22:47 -0700, Karl Williamson пишет:
>> On 12/04/2017 10:03 PM, Yaroslav Kuzmin wrote:
>>>
>>>
>>> Smoke log available https://drive.google.com/file/d/1CwVed6EgFWm29hbiN50NsNEGZGYw7vkP
>> This is looking like it's a problem with the new multi-concat. Attached
>> is a test script that is an instrumented version of one of the failing
>> tests. Please run it by hand and email the results.
>>
>> Say you save the attached file as 'foo'
>>
>> ./perl -I lib foo > foo.out 2>&1
>>
>
> Log available at https://drive.google.com/file/d/1lVQMTU4FwlGEqecOhMlIyfwiEkIaa7TP
>

I asked the multiconcat writer to look at this log, and he has come up
with a modified script o that we can drill down closer to the cause. It
is attached. Run it just like you did above.

Thanks

[Dave Mitchell]

On Mon, Dec 11, 2017 at 12:52:35PM +0000, Yaroslav Kuzmin wrote:
>
> Log availabel at https://drive.google.com/file/d/1BwOR2pbtYRTJ4TqWMh33FtD9PZahf1HM<https://drive.google.com/file/d/1BwOR2pbtYRTJ4TqWMh33FtD9PZahf1HM/view?usp=sharing>

Unfortunately that run didn't trigger the failure code which has heavy
instrumentation attached. It turns out that there are 5 places in the test
script that do an eval and might fail. The attached davem2 script
instruments all 5 evals.

I would be grateful if you could run it.

--
I thought I was wrong once, but I was mistaken.

[Karl Williamson]

It is likely that the reason the failure code was not triggered this
time is that it was changed somewhat by the addition of the
instrumentation code. My guess, without knowing how anything works
here, is that the mention of the variables being debugged causes them to
not be optimized away.

[Dave Mitchell]

On Tue, Dec 12, 2017 at 06:23:47AM +0000, Yaroslav Kuzmin wrote:
> Log available at https://drive.google.com/file/d/1jGDVWAw-L_ZBZHzfupDKy5uBNRKI9Dm7<https://drive.google.com/file/d/1jGDVWAw-L_ZBZHzfupDKy5uBNRKI9Dm7/view?usp=sharing>

Strange. In that run, it is failing four tests. In each case, a string
concatentation is inserting a spurious null character. For example with

$function = "XDIGIT";
$suffix = "_utf8"; # or "_LC_utf8"
$char = "\x{b3}"; utf8::upgrade($char);
$utf8_param_code = "0"; # or "-1"

"test_is${function}$suffix('$char',";

the result is

"test_isXDIGIT\x{0}_utf8('\x{b3}', 0)"

with similar results for the variants shown in the comments.

If you do further runs, do you get the same 4 tests consistently failing?
Can you run the attached test script and show any output. If should produce
no output normally.

--
My Dad used to say 'always fight fire with fire', which is probably why
he got thrown out of the fire brigade.

[Dave Mitchell]

On Thu, Dec 14, 2017 at 06:09:34AM +0000, Yaroslav Kuzmin wrote:
> No output :

This has me almost completely stumped.

What about the other thing I asked:

> If you do further runs, do you get the same 4 tests consistently failing?

That is, if you run the davem2 script multiple times, do you always get
these same test numbers failing:

$ grep '^not' davem2.py.log
not ok 135890 - isXDIGIT_utf8( \x{B3, MIDDLE DOT}, _safe ) didn't give error
not ok 135891 - isXDIGIT_utf8( \x{B3, MIDDLE DOT}, deprecated unsafe ) didn't give error
not ok 135892 - isXDIGIT_LC_utf8( \x{B3, MIDDLE DOT}, _safe ) (C locale) didn't give error
not ok 135894 - And got expected message
not ok 135895 - Got a single warning besides
not ok 135896 - isXDIGIT_LC_utf8( \x{B3, MIDDLE DOT}, deprecated unsafe ) (C locale) didn't give error

And if not, can you get consistent failures by setting the environment
variable PERL_HASH_SEED=0 ?

--
In economics, the exam questions are the same every year.
They just change the answers.