[4.2] SR-1464: NSNumber.description is not compatible between OS X and linux

[spevans]

• Use Swift's stringification of numbers for .description and .stringValue

• Use String(format:locale:...) with the correct format for the
  NSNumber type if locale is not nil.

• Enable use of __CFStringFormatLocalizedNumber() on
  DEPLOYMENT_TARGET_LINUX.

(cherry picked from commit 6450221)

• Add NSDecimalNumber.description

(cherry picked from commit 7cb608f)

[spevans]

@swift-ci test 4.2

[weissi]

@swift-ci test 4.2

[weissi]

@swift-ci test 4.2

[futurejones]

As reported in https://bugs.swift.org/browse/SR-9985 since the addition of this PR
TestNSNumber.test_descriptionWithLocale and TestNSString.test_initializeWithFormat3
are now failing on Linux/Ubuntu/AArch64

[weissi]

@futurejones thanks for reporting! So this PR passed all the tests here, are the ones you mention disabled?

Could this (on swift5) maybe be an artefact of vendoring ICU on Swift 5?

We definitely need to sort this out before releasing. @spevans / @kevints you know more about the ICU interactions here. Any insight?

[spevans]

Its odd that the swift-5.0-branch fails because none of the differences between 5.0 and master should matter. Also @futurejones tested 5.0 with #1866 which is a specific check for ICU >= 61, which is the only thing that could possibly make any difference so I cant see why that would fail either. Ubuntu18 comes with ICU60 so it must be getting built.

[futurejones]

@weissi These 2 tests TestNSNumber.test_descriptionWithLocale and TestNSString.test_initializeWithFormat3 were passing in 4.2.1 RELEASE and as far as I know they were passing in the 4.2.2 RELEASE. I am currently doing a 4.2.2 RELEASE build to verify this.
They are now failing in the current 4.2-branch. @spevans indicated that this PR may be the cause.
I have been testing on Ubuntu 16.04 only.

[futurejones]

To clarify these tests were passing on a build done from the swift-4.2-branch on 2019-02-21.

Test Case 'TestNSNumber.test_descriptionWithLocale' started at 2019-02-25 10:19:02.135
Test Case 'TestNSNumber.test_descriptionWithLocale' passed (0.0 seconds)

Test Case 'TestNSString.test_initializeWithFormat3' started at 2019-02-25 10:19:06.142
Test Case 'TestNSString.test_initializeWithFormat3' passed (0.0 seconds)

[spevans]

@weissi I think we should revert this one for now. I cant work out why it is breaking but it could just be an arm specific build issue. However the swift-5.0-branch breaking v master working is more worrying as I cant see any difference between the 2 branches wrt this change. But I think that would need to be looked at first.

[weissi]

thanks for looking into this @spevans .

[weissi]

@spevans I created #1950 to revert this

[spevans]

Further testing found that the fix for signed char in #1871 fixes this issue on ARM64 - the bug was that localised formatting was being disabled. So it should be ok to reintroduce after 4.2.3 has been released.