| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
| |
Bring changes from "trunk merging" branch back to this feature
branch, to make further work on the branch easier.
Tested on:
N/A - this branch is not compilable currently.
|
|
|
|
|
|
|
|
| |
Synchronize metadata journaling branch with changes from trunk.
Tested on:
Mac OS X/32 10.6.8 (amazon)
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bring metadata journaling branch closer to trunk by bringing back
updates to internal error reporting macros, routines that deal with writing
object header messages to the superblock extension, and some object header
query routines.
Also cleanups to journaling code & tests to use this code and to clean
up in general.
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in debug mode
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
| |
Rename H5AC_set() to H5AC_insert_entry()
Get rid of H5C_set_skip_flags() & related flags
Tested on:
Eyeballed only, very simple...
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Remove 'size' parameter from H5AC2_set()/H5C2_insert_entry(), to align
better with trunk. Use the 'image_len' cache client callback to retrieve the
size of an inserted entry. Also, add flags to the cache client class structure,
to indicate several client behaviors: speculative loads & compressed storage
on disk (which were previously dependent on the 'image_len' callback only
being used to detect size changes during load (deserialize) operations).
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in debug mode
Mac OS X/32 10.6.3 (amazon) in debug mode
Mac OS X/32 10.6.3 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add 'get_load_size' client callback to determine the image size for
reading a metadata cache entry from the file.
Remove the 'size' parameter from the H5[A]C_protect() calls.
General cleanups to align with code on trunk.
Tested on:
Mac OS X/32 10.6.3 (amazon) in debug mode
Mac OS X/32 10.6.3 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
| |
Simplify testing routines a bit more by dropping unused 'dirty_insert'
parameters. Also, protect global heap and resize it in the cache when
extending one.
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
Mac OS X/32 10.6.3 (amazon) in debug mode
Mac OS X/32 10.6.3 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
|
|
|
|
|
|
|
|
|
|
| |
Remove the size parameter from H5[A]C_unprotect() and the
H5[A]C__SIZE_CHANGED_FLAG, switching all clients & tests over to using
H5[A]C_resize_entry appropriately.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug & prod
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
| |
Correct test for resizing protected entries to check results correctly.
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
|
|
| |
Switch H5AC_resize_pinned_entry() to work on protected entries as well,
and rename to H5AC_resize_entry()
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
Mac OS X/32 10.6.3 (amazon) in debug mode
Mac OS X/32 10.6.3 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
|
|
|
|
|
|
|
|
| |
Rename H5[A]C_rename to H5[A]C_move_entry.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug & production
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
| |
Correct some sanity checks in test code.
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
| |
Rename H5[A]C_mark_pinned_or_protected_entry_dirty() to
H5[A]C_mark_entry_dirty() and get rid of H5[A]C_mark_pinned_entry_dirty().
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
Mac OS X/32 10.6.3 (amazon) in debug mode
Mac OS X/32 10.6.3 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
|
|
|
|
|
|
|
|
|
|
| |
Simplify parameters to 'deserialize' cache client callback.
Tested on:
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
Mac OS X/32 10.6.3 (amazon) in debug mode
Mac OS X/32 10.6.3 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
|
|
|
|
|
|
|
|
|
|
|
|
| |
More changes to move metadata journaling code closer to trunk:
- Retire old (H5AC1) cache code & tests
- Remove 'clear_dirty_bits' callback from client class struct
- Eliminate 'addr' & 'len' parameters from 'free_icr' client callbacks
Tested on:
Mac OS X/32 10.6.3 (amazon) debug & production
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
| |
Mostly minor code cleanups and reformatting to better set up udata
parameters when protecting metadata cache entries. Also, some changes to
better align with state of code on the trunk.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
| |
Fix a problem in the metadata accumulator code and clean up a few
compiler warnings.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
| |
Bring further into alignment with current state of trunk.
Tested on:
Mac OS X/32 (10.6.3) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
|
| |
Bring metadata journaling branch into closer correspondence with trunk:
extract data structure freeing routines from metadata cache client 'destroy'
callbacks.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
|
| |
Bring metadata journaling branch into closer correspondence with trunk:
extract data structure freeing routines from metadata cache client 'destroy'
callbacks.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
| |
Extract data structure 'destroy' routines from metadata cache client
'destroy' callbacks.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
| |
Extract data structure 'destroy' routines from metadata cache client
'destroy' callbacks.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
| |
Cleanups & renamings for extracted data structure 'destroy' routines.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug, production & parallel
(h5commit not required on this branch)
|
|
|
|
|
|
|
|
| |
Remove unused symbol table entry dirty flag.
Tested on:
Mac OS X/32 10.6.3 (amazon) w/debug, production & parallel
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
| |
More convergence of the metadata journaling branch toward to the trunk.
Tested on:
Mac OS X/32 10.6.2 (amazon) w/debug, prod & parallel
(h5committest not required on this branch)
|
|
|
|
|
|
|
| |
Revert Mac-related change (to disable aio_fsync() calls).
Tested on:
None - eyeballed only, too minor to require testing
|
|
|
|
|
|
|
|
|
|
| |
Finish renaming cache-related files & symbols on metadata journaling
branch. Metadata cache-related symbols/files with no suffix now have a '1'
suffix and symbols/files with a '2' suffix now have no suffix.
Tested on:
Mac OS X/32 10.6.2 (amazon) w/debug, prod & parallel
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
| |
Rename "old" cache code and routines from H5AC to H5AC1.
Tested on:
Mac OS X/32 10.6.2 (amazon) w/debug, prod & parallel
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Added code allowing async journal entry writes for metadata journaling,
along with associated test code.
At present, the code compiles and passes tests under Linux and Solaris.
code compiles under MacOS, but usually locks up the host machine under
test.
code fails to compile under BSD.
no other platforms tested.
|
|
|
|
|
|
|
|
|
|
| |
Diff reduction against v1 cache code, along with bringing in some very
minor changes to the v1 cache code that weren't brought over to the v2 cache
code.
Tested on:
Mac OS X/32 10.6.2 (amazon) debug & prod
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
| |
Small cleanups and move v1 cache maintenance macros into package header
file, to better align with changes to v2 cache.
Tested on:
Mac OS X/32 10.6.2 (amazon) debug & production
(h5committest not required on this branch)
|
|
|
|
|
|
|
| |
Remove trailing whitespace from source code files.
Tested on:
None - just eyeballed
|
|
|
|
|
|
|
|
|
|
|
|
| |
Code cleanups for better alignment with trunk, along with style issues,
etc.
Also, get journal test files to work correctly when using a srcdir
build.
Tested on:
Mac OS X/32 10.6.2 (amazon) w/debug, production & parallel
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
| |
Minor brush-clearing changes to converge metadata_journaling branch and
trunk.
Tested on:
Mac OS X/32 10.6.2 (amazon) debug & prod
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
| |
Various tweaks, upgrades & corrections from merging the local heap
changes on this branch back to the trunk.
Tested on:
Mac OS X/32 10.6.2 (amazon) debug & prod
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
|
| |
Slush changes back & forth between trunk, the merging branch and the
metadata journaling branch to level them out to a reasonably common set of code
to work from for the next set of more significant changes.
Tested on:
Mac OS X/32 10.6.2 (amazon) w/debug
(h5committest not required on this branch)
|
|
|
|
|
|
|
| |
Whitespace cleanup.
Tested on:
None - eyeballed only.
|
|
|
|
|
|
|
|
|
| |
Code cleanups to remove compile warnings and improve the style of
changes.
Tested on:
Mac OS X/32 10.6.2 (amazon) debug & production
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
| |
Code cleanups & tweaks to converge branch toward trunk.
Tested on:
Mac OS X/32 10.6.2 (amazon) debug & production
(h5committest not required on this branch)
|
|
|
|
| |
Ran chkmanifest on jam.
|
|
|
|
|
|
|
|
| |
manifest --
forgot to do this on my last checkin.
Testing -- ran bin/chkmanifest
|
|
|
|
|
|
|
|
|
| |
Removed a bunch of debug code.
Other minor changes.
Did only limited testing -- just build and test on Phoenix (AMD64 linux
with debug).
|
|
|
|
|
|
| |
fixing that now
tested by running chkmanifest
|
|
|
|
| |
Tested only on Phoenix -- serial source dir and in place builds.
|
|
|
|
|
|
|
|
|
|
| |
Clean up code, remove unused variables, remove "naked" printf()s, make
h5recover test script work in 'srcdir' build, etc. (The h5recover tests are
failing still and the script prints "PASSED" when it doesn't, but we'll work on
that more next)
Tested on:
Mac OS X/32 10.5.6 (amazon)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
recognition
and rejection of attempts to apply the wrong journal file to a corrupt
HDF5 file. Specifically, I made the following changes:
1) Moved all journaling data into the journaling in progress superblock
extension message.
2) Added a "magic number" to the journaling in progress message, with
the same "magic number being added to the header of the associated
journal file.
3) Modifications to library test code to support the above.
4) Modified h5recover to examine the supplied hdf5 file, determine if
it is in fact a HDF5 file, if so determine if it is marked as having
journaling in progress, and if it does, extract the contents of the
journaling in progress super block extension message.
5) Modified h5recover to examine the supplied journal file, determine
if it is in fact a HDF5 journal file, and if so, extract the data
from its header.
6) Modified h5recover to refuse to apply the supplied journal file to
the supplied HDF5 file unless the "magic numbers" obtained from these
files matches.
7) Added an examine option to h5recover that causes it to examine and
report on the supplied files, but do nothing. This option exists
primarily to facilitate testing, but I expect that some users will
find it useful as well.
8) Added test code to exercise items 4-7. Note that while I have tried
to cover the more likely cases, this test code is extremely cursory.
In particular, the code to examine the supplied HDF5 file is barely
tested at all. Need a library of HDF5 files exibiting the full range
of possible super block and super block extension message structures
to test this properly.
9) In passing, tighened up the code that controls dumps of "possibly
significant" differences between the contents of the control and
recovered data sets in the h5recover. It should now ignore one
integer matches in what appears to be garbage raw data.
Tested: serial and parallel on Phoenix
serial and parallel on Jam
serial on Linew
serial on Liberty
All tests were done in debug mode.
|
|
|
|
|
| |
Tested only on Phoenix (serial Linux AMD64) as changes were minor, and
any errors should be caught in the next checkin.
|
|
|
|
|
|
| |
prior to the
last checkin.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
driver in
metadata cache tests.
1) Fixed the core file driver failures previously observed on some
platforms when running the cache2 tests with the core file driver
enabled. This was done by allocating all needed memory on file open.
2) Added code to cache2.c to allow the use of the core file driver to
be forced via the HDF5_DRIVER environment variable.
3) Added code to try to figure out whether using the core file driver
in cache2 makes sense, and then use it or not as seems appropriate
unless overridden via the HDF5_DRIVER environment variable.
This code only works under Linux and BSD (including MacOS). For now
at least, we use regular files in all other cases unless directed
otherwise.
Note that this required a fair bit of configuration code massage.
4) Updated Makefile.am in examples to run the new mdj_api_example.c
example. Forgot to "svn add mdj_api_example.c" before this checkin,
but will do so shortly.
Tested on:
Duty (serial), Liberty (serial), tejeda (serial), jam (serial and
parallel), and Phoenix (serial and parallel). Note that Phoenix is
now 64 bit AMD64 Linux.
|
|
|
|
| |
metadata journaling branch
|