Create a file with permissions that respects umask

[niw]

Problem

Data.write(to:) is a only method in the Foundation that can create a
regular file.
However, it ignores umask and always set 0600 permission unlike
macOS Foundation, which respects process umask.

Solution

1. With .atomic write option

   It uses mkstemp(3) in _NSCreateTemporaryFile, which is always
   creating a file with 0600 permission, if the system follows
   the latest POSIX specification or the permission is undefined.

   On macOS Foundation, therefore _NSCreateTemporaryFile uses
   mktemp(3) and open(2) instead to respect umask.

2. Without .atomic write option

   It uses 0o600 even if it uses open(2) that respects umask.
   Simply gives 0o666 instead.

This is a bug caused by previous commit in
#1876.

Swift JIRA is https://bugs.swift.org/browse/SR-13307.

NOTE: This patch addressed the issue on #2854, which reverts
original pull request #2851.

[spevans]

@swift-ci test

[millenomi]

It looks like this was just heavy traffic triggering the type evaluation timeout.

[millenomi]

@swift-ci please test

[niw]

/home/buildnode/jenkins/workspace/swift-corelibs-foundation-PR-Linux/swift-corelibs-foundation/Sources/Foundation/NSData.swift:464:30: error: the compiler is unable to type-check this expression in reasonable time; try breaking up the expression into distinct sub-expressions
16:20:59             let createMode = Int(S_IREAD | S_IWRITE | S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH)
16:20:59                              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Hmm, intersting...

[niw]

Oh, no, S_IREAD | S_IWRITE shouldn't be here. My bad.
Also now I understand that each flag are probably literal on Linux and that's why it takes long time to type checking.
Addressed these issues.

[millenomi]

@swift-ci please test

[niw]

/home/buildnode/jenkins/workspace/swift-corelibs-foundation-PR-Linux@2/swift-corelibs-foundation/Sources/Foundation/NSData.swift:468:34: error: ambiguous use of 'S_IRUSR'
11:31:40             let createMode = Int(S_IRUSR) | Int(S_IWUSR) | Int(S_IRGRP) | Int(S_IWGRP) | Int(S_IROTH) | Int(S_IWOTH)
11:31:40                                  ^

Ugh, multi-platform change isn't easy really.

[spevans]

@swift-ci please test

[spevans]

@swift-ci please test

[niw]

05:02:09 Test Suite 'JobExecutorTests' started at 2020-08-29 12:02:09.126
05:02:09 Test Case 'JobExecutorTests.testStubProcessProtocol' started at 2020-08-29 12:02:09.126
05:02:09 Exited with signal code 4

03:59:39 ******************** TEST 'Swift-Unit :: ClangImporter/./SwiftClangImporterTests/ClangImporterTest.missingSubmodule' FAILED ********************
03:59:39 Note: Google Test filter = ClangImporterTest.missingSubmodule
03:59:39 [==========] Running 1 test from 1 test case.
03:59:39 [----------] Global test environment set-up.
03:59:39 [----------] 1 test from ClangImporterTest
03:59:39 [ RUN      ] ClangImporterTest.missingSubmodule
03:59:39 /Users/buildnode/jenkins/workspace/swift-corelibs-foundation-PR-macOS/branch-master/swift/unittests/ClangImporter/ClangImporterTests.cpp:194: Failure
03:59:39 Value of: context->getModule({ Located<Identifier>{ CLib2, {} } }) == nullptr
03:59:39   Actual: false
03:59:39 Expected: true
03:59:39 [  FAILED  ] ClangImporterTest.missingSubmodule (29 ms)
03:59:39 [----------] 1 test from ClangImporterTest (29 ms total)
03:59:39 
03:59:39 [----------] Global test environment tear-down
03:59:39 [==========] 1 test from 1 test case ran. (29 ms total)
03:59:39 [  PASSED  ] 0 tests.
03:59:39 [  FAILED  ] 1 test, listed below:
03:59:39 [  FAILED  ] ClangImporterTest.missingSubmodule
03:59:39 
03:59:39  1 FAILED TEST

Hmm... both CI failed on different position and... seems unrelated to the patch itself...? 🤔

[spevans]

@swift-ci please test