This test assumed that a vector of 4 ints needed 16-byte alignment, but
this is target specific. Now check that the runtime provides at least
as strong alignment as the compiler expects.
We now correctly handle ivars that overlap with the end of what the
compiler thinks is the start of the superclass and bitfields at the
start of a class.
When removing a hash entry that was in the cell assigned to its hash, we
would clear the first bit in the second maps. If this entry was a
secondary value with the same hash, then this value became unreachable.
This very rarely showed up for two reasons. First, most of the tables
are insert-only and so we never try to remove things from them. Second,
it requires a particular sequence of inserts. It occasionally caused
weak references to be susceptible to use after free.
This cleans up handling of objects that are not reference counted and
makes their interactions with ARC more consistent. We should probably
generalise this somewhat - it currently special cases NSConstantString
and NSGlobalBlock, but it would be nice to have an API for constant
objects.
This has no effect most of the time, but if the runtime is built with
llvm-cov support, then this will cause each test to emit a .profraw
file, which can be merged with:
$ llvm-profdata merge -sparse Test/*.profraw -o libobjc.profdata
This can then be used to generate coverage reports from the test suite
and find code paths that are not currently being tested.
Coverage checking of the test suite showed that objc_disposeClassPair
wasn't tested at all, which then led to discovering that it didn't
unregister the class.
This test was accidentally passing sometimes, with the isa pointer being
set using some bit of memory in inter-object padding. This breaks
horribly with an allocator that packs objects densely.
Methods now include a selector and extended type encoding, rather than a
method name and lgacy type encoding. Older ones are auto-upgraded.
Expose the extended type encoding via a function that JavaScriptCore
expects to exist.
This change set incorporates a number of changes that all needed to
happen together:
* The imp is now the first field of the `objc_method` structure. This
makes it possible to extend the structure without breaking anything
that relies on being able to access the IMP.
* There is no owner in the slot, so we must use other mechanisms for
determining the owner of a method (e.g. whether the same method appears
in the superclass)
* Again, because there is no owner in the slot, we can't use this as a
fast path for finding the C++ construct / destruct methods. These are
now cached in the class structure when they are found.
* The version field is gone from the slot and now we provide a global
version. This is based on the observation that method replacements
are relatively infrequent and the overhead of invalidating all method
caches is cheaper than adding extra state for every (class, method)
pair.
* A number of the runtime functions are simplified because replacing
the IMP in a `Method` now implicitly updates the dtable.
With the old ABI, we generated a copy of every protocol in every
compilation unit that declared it. We now emit protocols only if they
are referenced (and have the linker deduplicate them).
This test was previously failing with the Apple runtime.
Without this fix, we would lose associated objects silently after adding
the 11th. We would also allocate full pages for each object after the
11th because we couldn't find empty slots.
David Chisnall
David Chisnall
Dustin Howett
davidchisnall
David Chisnall