Brings the thread-safety recursive writer locks to 1.12 (#466)

* First cut at replaceing the existing mutex with a recursive R/W lock.

This implementation has the following issues:

    1) pthreads implementation only -- we still need a windows version.

    2) must investigate thread cancelation issues

    3) Error reporting is very poor.  I followed the error reporting on
       the existing thread safe code, but this should be re-visited and
       improved.

Code is currently setup to use the new recursive R/W lock instead of
the global mutex to control entry to the library in threadsafe builds.
To revert to the global mutex, set H5TS__USE_REC_RW_LOCK_FOR_GLOBAL_MUTEX
in H5TSprivate.h to FALSE.

Added a reasonably robust regression test for the reursive R/W lock
in test/ttsafe_rec_rw_lock.c

Note that the change to hl/src/H5LTanalyse.c is an artifact of clang-format.

Tested serial threadsafe debug and production on jelly, and also regular
serial / debug.

On Windows builds, the new recursive R/W lock should not be built and
we should use the existing global mutex -- however this is not tested
at this time.

* Updates CMake to build recursive RW lock test

* Updates to allow building on Windows

* Moves #if statements to better protect non-RW lock code

* Adds configure and CMake options for the recursive RW locks

* Committing clang-format changes

* Updates RELEASE.txt and the build options

* Renames H5TS RW lock things

* Makes struct members platform independent

Also removes _ptr from identifiers

* Partial thread-safety RW locks platform independence

* Committing clang-format changes

* Pthreads side of things is platform-independent now

* Formatted source

* Added Windows equivalents for some Pthreads calls

* Rename H5TS takedown call to destroy

* Reorg of RW lock code

* Committing clang-format changes

* Changes to Pthreads code after development on Visual Studio

* Moves stats macros to static inline functions and tidies memory allocs

* Converts RW lock print stats call to use C99 formatting

* Fixes typos

* Formatted source

* Updates the RELEASE.txt note to indicate no Win32 threads support

Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>

1 files changed, 25 insertions, 0 deletions

diff --git a/release_docs/RELEASE.txt b/release_docs/RELEASE.txt
index 807d3ab..66b78ac 100644
--- a/release_docs/RELEASE.txt
+++ b/release_docs/RELEASE.txt
@@ -47,6 +47,31 @@ New Features
 
     Configuration:
     -------------
+    - Added an option to make the global thread-safe lock a recursive R/W lock
+
+        Prior to this release, the HDF5 library supported multi-threaded
+        applications by placing a recursive global lock on the entire library,
+        thus allowing only one thread into the library at a time.
+
+        While this is still the default, the library can now be built with the
+        recursive global lock replaced with a recursive read / write (R/W) lock
+        that allows recursive writer locks.
+
+        Currently, this change results in no functional change in the HDF5
+        library, as all threads will have to acquire a write lock on entry, and
+        thus obtain exclusive access to the HDF5 library as before. However, the
+        addition of the recursive R/W lock is a prerequisite for further work
+        directed at allowing some subset of the HDF5 API calls to enter the
+        library with read locks.
+
+        CMake:      HDF5_USE_RECURSIVE_RW_LOCKS (default: OFF, advanced)
+
+        Autotools:  --enable-recursive-rw-locks [default=no]
+
+        This feature only works with Pthreads. Win32 threads are not supported.
+
+      (DER - 2021/05/10)
+
     - CMake no longer builds the C++ library by default
 
       HDF5_BUILD_CPP_LIB now defaults to OFF, which is in line with the