[ruby-core:81649] [Ruby trunk Bug#13644] Windows - Setting Time.now

[Greg.mpls@gmail.com]

Issue #13644 has been updated by MSP-Greg (Greg L).

File test-logger-test_logdevice.rb.patch added

nobu,

Attached is the patch I'm using in my MinGW build system for the test in question.  It passes, and my [test-all](https://msp-greg.github.io/file.mingw_test-all.html#test-all-3-failures) is now at 3 failures.

I ran into three issues with the current test mock.

1. Once I patched logger to use atime, the test wouldn't pass if the FakeTime.now DST setting was different that the current system DST setting.  So, I moved the mock FakeTime.now to today.
2. Temp Directories using a block.  I had issues with readline tests and temp files.  With this, I could not get all the asserts to pass with a block.  As to files, I believe Windows may have issues with - if process A does not close a file, process B can neither close nor delete the file.
3. As I understand Logger, there was code that would not be used in a normal Logger application.  Removed/changed code that opened the log file first, code that closed and then opened (with `.new`) the log file, etc.

The current test uses a midnight delta of -1mS and +01:01:01.  The plus delta seems high, and from the [GH 539](https://github.com/ruby/ruby/pull/539), the author (megayu) states

> In my real situation, if there is some events in 23:59:59, it will never shift the log file.

Seems to imply a minus delta of around a second.  So, I set up the test with a symmetrical delta.  I tested three deltas, 1000mS, 500mS, and 10mS, no failures with three processes running 50 tests each.

I can't test any other platform without pushing a PR.  I learned a bit about the logger library.

Aside - found an interesting line in the file #729, just after the test in question.

```
env_tz_works = /linux|darwin|freebsd/ =~ RUBY_PLATFORM # borrow from test/ruby/test_time_tz.rb
```

----------------------------------------
Bug #13644: Windows - Setting Time.now
https://bugs.ruby-lang.org/issues/13644#change-65345

* Author: MSP-Greg (Greg L)
* Status: Open
* Priority: Normal
* Assignee: 
* Target version: 
* ruby -v: ruby 2.5.0dev (2017-06-04 trunk 59013) [x64-mingw32]
* Backport: 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN
----------------------------------------
While reviewing a MinGW build `test-all` failure in [TestLogDevice#test_shifting_midnight_exist_file](https://github.com/ruby/ruby/blob/trunk/test/logger/test_logdevice.rb#L687-L727), I determined the cause.  The test in question (and other tests that seem to have newer/better exception handling) sets `Time.now`.

The thread [StackOverflow - Change system date programmatically](https://stackoverflow.com/questions/650849/change-system-date-programmatically) seems to imply that in some instances/configurations, changing the system time on Windows systems requires elevated permissions.

I tested on both a mswin build (`ruby 2.5.0dev (2017-05-27 trunk 58922) [x64-mswin64_140]`) and a MinGW build (`ruby 2.5.0dev (2017-06-08 trunk 59046) [x64-mingw32]`), and both responded to:

```ruby
Time.now = Time.mktime(2017, 1, 3, 1, 1, 1)
```

with the following eror:

```
undefined method `now=' for Time:Class (NoMethodError)
```

Both also had `false = Time.respond_to?(:now=)`.

So, I'm confused as to how this test passes on mswin.  Regardless, the test requires a skip to bypass it and allow MinGW test-all to pass.

I thought I'd file an issue before doing a PR.  I'm building and testing on Win7.  Finally, if anyone has time, where is the method `now=` defined?

---Files--------------------------------
test-logger-test_logdevice.rb.patch (3.6 KB)


-- 
https://bugs.ruby-lang.org/