[foreign] test_time timezone trouble (TZ)

[Jorn Vernee]

Hey Jon,

One of the other panama guys suggested to just drop the localtime() test 
and call gmtime() instead, which does not have the environment variable 
dependency. We currently have the luxury to do that, but the future 
might be different, so thanks for the RFE.

FWIW the file in question [1] is using the '@run testng' action tag, so 
the option would need to be added there as well I guess.

Jorn

[1] : 
http://hg.openjdk.java.net/panama/dev/file/01ecea66dd99/test/jdk/java/foreign/StdLibTest.java

Jonathan Gibbons schreef op 2018-09-22 01:30:
> Jorn,
> 
> Yes, short term, I would recommend option 2 as the way to go. There
> should be examples to copy, and if you want to get cute and avoid
> another source file, your test could have a main program that checks
> if TZ is set, and if so, it re-execs itself without TZ.
> 
> Long term, I think the right way to handle this would be either as an
> option or an argument on the "@run main" and "@run shell" action tags.
> 
> Neither are great, syntax-wise, but either would be good enough for
> your specific case:
> 
> For example,
> 
> main[/fail][/manual][/othervm][/policy=<file>][/secure=<class>]
> [/timeout=<seconds>][/e:NAME=VALUE]* <vm-opt>* <class> <arg>*
> 
> main[/fail][/manual][/othervm][/policy=<file>][/secure=<class>]
> [/timeout=<seconds>] [NAME=VALUE]* <vm-opt>* <class> <arg>*
> 
> I'll file an RFE for jtreg.
> 
> -- Jon
> 
> On 09/21/2018 04:18 PM, Jorn Vernee wrote:
> 
>> Hey Jon,
>> 
>> Thanks for the suggestions (and thank you Michel for forwarding!).
>> 
>> I will go with option 2 I think. I'm running the tests through `make
>> test`, so I'm not sure how I would use the command-line flag, but
>> setting a flag all the time is just what I'm trying to avoid any
>> ways :) It doesn't look like there is a way to pass jtreg options
>> from the test file itself? (maybe something like `@params "-eTZ="`).
>> 
>> 
>> The test is calling a C standard library function, so that is to
>> blame for the environment variable dependency.
>> 
>> Thanks,
>> Jorn
>> 
>> Jonathan Gibbons schreef op 2018-09-21 23:29:
>> Jorn,
>> 
>> There is only limited amount of support for setting env variables in
>> 
>> jtreg tests, because at least in the
>> early releases of JDK, the use of environment variables was
>> discouraged.
>> 
>> Using `@run shell unsettz.sh` will not have the desired effect
>> because
>> it will only affect the shell
>> that is created to run your shell script, and will not have a
>> persistent environment on any other
>> process.
>> 
>> If you don't mind always setting a command-line option, you might
>> try
>> using `-eTZ=` on the jtreg
>> command-line, which will unset TZ for all tests, although it would
>> be
>> more accurate to say that
>> it will not set it for any tests.
>> 
>> The only way to change the setting of TZ for any specific test will
>> be
>> to do one of the
>> following:
>> 
>> 1. Write the test as a shell script that unsets TZ and then runs any
>> 
>> Java code, using various
>> shell environment variables that will be available; these
>> generally begin with "TEST".
>> 
>> 2. Write the test as a Java program that execs a child process to
>> run
>> Java, with the exact set
>> of environment variables you want to pass in. There are system
>> properties that identify
>> the test JDK, test class path and so on; these generally begin
>> with "test.".
>> 
>> These days, writing shell tests is generally discouraged, because it
>> 
>> is notoriously difficult to get it
>> right on all relevant platforms, and because these days, it is
>> reasonably easy to use either
>> the Java SE Process[Builder] API or test library API to invoke java.
>> 
>> 
>> Hope that helps; if you have follow-up questions, please cc: me
>> directly or use one of the jtreg-* aliases,
>> since I am not on the panama-dev list.
>> 
>> -- Jon
>> 
>> On 09/21/2018 02:12 PM, Michel Trudeau wrote:
>> [adding jtreg mailing list to seek out answer about jtreg and TZ on
>> Windows]
>> 
>> On Sep 21, 2018, at 12:44 PM, Jorn Vernee <jbvernee at xs4all.nl>
>> wrote:
>> 
>> Hello guys,
>> 
>> I was running into a problem with the test_time test in StdLibTest.
>> The generator for test values was apparently generating invalid
>> values. After fixing that, all the test iterations were failing
>> because the hours were off by one (the output of
>> LocalDateTime.getHours() is being compared to the output of
>> localtime().hour()). Good thing somebody else on the internet seems
>> to have had the same problem [1]. TL;DR when using the MSVC version
>> of localtime, and when running in cygwin, the function tries to
>> interpret the TZ environment variable, but since that has a unix
>> format (courtesy of cygwin), the interpretation fails and defaults
>> to GMT. When TZ is not set, it defaults to the system default
>> timezone, which is also what's being tested against.
>> 
>> I can get the tests to pass by using `unset TZ` in the cygwin
>> terminal before running them, but I'd have to do that every time I
>> reboot it. I was trying to unset TZ automatically by using jtreq
>> `@run shell unsettz.sh` where unsettz.sh is a file containing just
>> the command `unset TZ`. It seems to be running successfully
>> according to the test output:
>> 
>> ACTION: shell -- Passed. Execution successful
>> REASON: User specified action: run shell unsettz.sh
>> TIME:   0.126 seconds
>> messages:
>> command: shell unsettz.sh
>> reason: User specified action: run shell unsettz.sh
>> elapsed time (seconds): 0.126
>> STDOUT:
>> STDERR:
>> 
>> But it doesn't seem to affect the tests itself, and they still fail
>> (still off by 1 hour).
>> 
>> I was wondering if there is a way to let jtreg control environment
>> variables? Or maybe you can suggest a different solution?
>> 
>> The jtreg guide [2] mentions that TZ will be propagated from Windows
>> 'if set', but I need it to be not set, or automatically set to the
>> system's default time zone (by default it's blank).
>> 
>> (other than that, tests are looking good: `passed: 24; failed: 4;
>> error: 1`. I just need to fix structs by value, which on Windows
>> cheats and just passes a pointer. 2 failing tests are from jextract
>> missing)
>> 
>> Thanks,
>> Jorn
>> 
>> [1] : https://stackoverflow.com/q/11655003
>> [2] : http://openjdk.java.net/jtreg/tag-spec.html