This function mixes the bits in the hash value, which improves Dictionary
performance for keys with bad hashes.
PrecommitBenchmark changes with greater than 7% difference:
``````````Dictionary2`,```1456.00`,```1508.00`,```1502.00`,````624.00`,````607.00`,````592.00`,`864.00`,``145.9%
``````````Dictionary3`,```1379.00`,```1439.00`,```1408.00`,````585.00`,````567.00`,````552.00`,`827.00`,``149.8%
````````````Histogram`,````850.00`,````849.00`,````851.00`,```1053.00`,```1049.00`,```1048.00`,`199.00`,``-19.0%
````````````````Prims`,```1999.00`,```2005.00`,```2018.00`,```1734.00`,```1689.00`,```1701.00`,`310.00`,```18.4%
``````````StrSplitter`,```2365.00`,```2334.00`,```2316.00`,```1979.00`,```1997.00`,```2000.00`,`337.00`,```17.0%
```````````````TwoSum`,```1551.00`,```1568.00`,```1556.00`,```1771.00`,```1741.00`,```1716.00`,`165.00`,```-9.6%
Regressions are in benchmarks that use `Int` as dictionary key: we are just
doing more work than previously (hashing an `Int` was an identity function).
rdar://17962402
Swift SVN r21142
The test harness now can recover after test crashes, allowing:
- check for crashes themselves (without reporting them to the Python lit driver,
which is about 10x slower -- even if CrashTracer is disabled);
- recover from unexpected test crashes and run the rest of the tests;
- this lays the groundwork for assertions that end the test execution, but
allow the rest of the tests to run (rdar://17906801).
Note that we don't spawn a fresh process for every test. We create a child
process and reuse it until it crashes.
Swift SVN r21090
To limit user confusion when using conditional expressions of type Bool?, we've decided to remove the BooleanType (aka "LogicValue") conformance from optional types. (If users would like to use an expression of type Bool? as a conditional, they'll need to check against nil.)
Note: This change effectively regresses the "case is" pattern over types, since it currently demands a BooleanType conformance. I've filed rdar://problem/17791533 to track reinstating it if necessary.
Swift SVN r20637
normalization
There is still some obscure bug with != on NSString, probably caused by
an ill-thought overload somewhere.
Part of rdar://17498444
Swift SVN r20518
normalization
There is still some obscure bug with != on NSString, probably caused by
an ill-thought overload somewhere.
Part of rdar://17498444
Swift SVN r20495
+= only extends arrays with another sequence of the same element type.
Fixes <rdar://problem/17151420> The use of the overloaded += operator in
Swift is inconsistent and confusing with Arrays.
Note that this commits generated 3 new radars against the type checker:
<rdar://problem/17751308>
<rdar://problem/17750582>
<rdar://problem/17751359>
Swift SVN r20274
Mechanically add "Type" to the end of any protocol names that don't end
in "Type," "ible," or "able." Also, drop "Type" from the end of any
associated type names, except for those of the *LiteralConvertible
protocols.
There are obvious improvements to make in some of these names, which can
be handled with separate commits.
Fixes <rdar://problem/17165920> Protocols `Integer` etc should get
uglier names.
Swift SVN r19883
...unless the type has less accessibility than the protocol, in which case
they must be as accessible as the type.
This restriction applies even with access control checking disabled, but
shouldn't affect any decls not already marked with access control modifiers.
Swift SVN r19382
- Follow LLVM conventions for emacs mode specification
- Use local variables suffix to make the output read-only (at least on
Emacs)
- But drop the admonitions not to edit the generated files;
line-directive mostly takes care of that problem now.
Swift SVN r19381
As before, there may be more things marked @public than we actually want
public. Judicious use of the frontend option -disable-access-control may
help reduce the public surface area of the stdlib.
Swift SVN r19353
If underlying NSString contained isolated surrogates, then we were crashing in
following ways:
- subscripting by index could crash;
- index pointing to the second code unit sequence was not moved backwards
correctly. Instead of moving it to pointing to the beginning of the view it
could be moved to point to the code unit before the beginning of the view.
Swift SVN r19230
implementation:
- don't crash on NSStrings that contain isolated surrogates (partial fix, there
is a different code path that dispatches CoreFoundation and it still
crashes);
- insert U+FFFD according to Unicode recommendation;
- fix decoding of surrogate pairs when the internal buffer is almost full
(we used to drop characters in that case, rdar://16833733 + dups).
Swift SVN r19147
Keep calm: remember that the standard library has many more public exports
than the average target, and that this contains ALL of them at once.
I also deliberately tried to tag nearly every top-level decl, even if that
was just to explicitly mark things @internal, to make sure I didn't miss
something.
This does export more than we might want to, mostly for protocol conformance
reasons, along with our simple-but-limiting typealias rule. I tried to also
mark things private where possible, but it's really going to be up to the
standard library owners to get this right. This is also only validated
against top-level access control; I haven't fully tested against member-level
access control yet, and none of our semantic restrictions are in place.
Along the way I also noticed bits of stdlib cruft; to keep this patch
understandable, I didn't change any of them.
Swift SVN r19145
In UTF-8 decoder:
- implement U+FFFD insertion according to the recommendation given in the
Unicode spec. This required changing the decoder to become stateful, which
significantly increased complexity due to the need to maintain an internal
buffer.
- reject invalid code unit sequences properly instead of crashing rdar://16767868
- reject overlong sequences rdar://16767911
In stdlib:
- change APIs that assume that UTF decoding can never fail to account for
possibility of errors
- fix a bug in UnicodeScalarView that could cause a crash during backward
iteration if U+8000 is present in the string
- allow noncharacters in UnicodeScalar. They are explicitly allowed in the
definition of "Unicode scalar" in the specification. Disallowing noncharacters
in UnicodeScalar prevents actually using these scalar values as internal
special values during string processing, which is exactly the reason why they
are reserved in the first place.
- fix a crash in String.fromCString() that could happen if it was passed a null
pointer
In Lexer:
- allow noncharacters in string literals. These Unicode scalar values are not
allowed to be exchanged externally, but it is totally reasonable to have them
in literals as long as they don't escape the program. For example, using
U+FFFF as a delimiter and then calling str.split("\uffff") is completely
reasonable.
This is a lot of changes in a single commit; the primary reason why they are
lumped together is the need to change stdlib APIs to account for the
possibility of UTF decoding failure, and this has long-reaching effects
throughout stdlib where these APIs are used.
Swift SVN r19045
