* Fixed memory bug: properly root repeated/map field when assigning.
Previously the protobuf extension would not properly root
memory from a repeated field or map when assigning to a
message field (see the attached test case). This could cause
crashes if the repeated field is subsequently accessed.
* Add accidentally-deleted Ruby test.
* Some more updates to PHP testing infrastructure (#8576)
* WIP.
* Added build config for all of the tests.
* Use ../src/protoc if it is available, for cases where Bazel isn't available.
* Added test_php.sh.
* Fix for the broken macOS tests.
* Move all jobs to use php80 instead of lots of separate jobs.
* Only pass -t flag if we are running in a terminal.
* Updated php_all job to use new Docker stuff.
* Fixed PHP memory leaks and arginfo errors (#8614)
* Fixed a bunch of incorrect arginfo and a few incorrect error messages.
* Passes mem check test with no leaks!
* WIP.
* Fix build warning that was causing Bazel build to fail.
* Added compatibility code for PHP <8.0.
* Added test_valgrind target and made tests Valgrind-clean.
* Updated Valgrind test to fail if memory leaks are detected.
* Removed intermediate shell script so commands are easier to cut, paste, and modify.
* Passing all Valgrind tests!
* Hoist addref into ObjCache_Get().
* Removed special case of map descriptors by keying object map on upb_msgdef.
* Removed all remaining RETURN_ZVAL() macros.
* Removed all explicit reference add/del operations.
* Added REFCOUNTING.md to Makefile.am.
* Updated upb version and fixed PHP to not get unset message field. (#8621)
* Updated upb version and fixed PHP to not get unset message field.
* Updated changelog.
* Fixed preproc test to handle old versions of Clang withot __has_attribute().
* A second try at fixing __has_attribute().
* Copy __has_attribute() fix to cc file also.
* Updated failure list for PHP for fixed test.
* Updated version of upb for Ruby (#8624)
* Updated upb.
* Preserve legacy behavior for unset messages.
* Updated failure list.
* Updated CHANGES.txt.
* Added erroneously-deleted test file.
* Fixed condition on compatibility code.
* Re-introduced deleted file again, and fixed Rakefile to not delete it.
* Fix generation of test protos.
* fix(ruby): Fix crash when calculating Message hash values on 64-bit Windows
* Better mapping for values outside the fixnum range
* Simpler downcasting of hash values
* Fix precedence
* Fix bundle on Ruby 2.4
* fix(ruby): Fix various exceptions in Ruby on 64-bit Windows.
Activates the secondary ObjectCache map on this platform, to prevent weak keys from being garbage collected. This happened on 64-bit Windows because pointers don't necessarily fit in a Fixnum, and were being represented as GC-able Bignums on that platform.
* Removed extraneous code, and used VALUE instead of intptr_t
* Call the C function for new object instance rather than evaling a Ruby string
* Select appropriate protoc binary for gem build
This should allow gem installation when
`bundle config force_ruby_platform true` has been set.
It has the unfortunate side effect of printing the protoc version during
build, but a quiet cross-platform solution is more complicated.
* Add path to wyhash header for source gem compilation
In our free() method, we were freeing the memory from the
upb arena but we were failing to free the memory for the
Ruby arena object. This was causing every Ruby arena object
to leak: even though the objects were getting GC'd, the
underlying memory was not getting released.
We now hold the mutex for both map insertions, to protect
against a concurrent GC that removes from the seconary map
before we can insert into the weak map.
This commit removes an unused reference and registers globals with the
GC so that they will never die. Ruby is getting a compacting GC, and it
means that these references can move. Registering them with
`rb_gc_register_mark_object` will ensure the constants don't move and
will not be collected.
Ruby <2.7 does not allow non-finalizable objects to be WeakMap
keys: https://bugs.ruby-lang.org/issues/16035
We work around this by using a secondary map for Ruby <2.7 which
maps the non-finalizable integer to a distinct object.
For now we accept that the entries in the secondary map wil never
be collected. If this becomes a problem we can perform a GC pass
every so often that looks at the contents of the object cache to
decide what can be deleted from the secondary map.
