Filter support has three goals:
1. Use the existing HDF5 filter implementations,
2. Allow filter metadata to be stored in the NumCodecs metadata format used by Zarr,
3. Allow filters to be used even when HDF5 is disabled
Detailed usage directions are define in docs/filters.md.
For now, the existing filter API is left in place. So filters
are defined using ''nc_def_var_filter'' using the HDF5 style
where the id and parameters are unsigned integers.
This is a big change since filters affect many parts of the code.
In the following, the terms "compressor" and "filter" and "codec" are generally
used synonomously.
### Filter-Related Changes:
* In order to support dynamic loading of shared filter libraries, a new library was added in the libncpoco directory; it helps to isolate dynamic loading across multiple platforms.
* Provide a json parsing library for use by plugins; this is created by merging libdispatch/ncjson.c with include/ncjson.h.
* Add a new _Codecs attribute to allow clients to see what codecs are being used; let ncdump -s print it out.
* Provide special headers to help support compilation of HDF5 filters when HDF5 is not enabled: netcdf_filter_hdf5_build.h and netcdf_filter_build.h.
* Add a number of new test to test the new nczarr filters.
* Let ncgen parse _Codecs attribute, although it is ignored.
### Plugin directory changes:
* Add support for the Blosc compressor; this is essential because it is the most common compressor used in Zarr datasets. This also necessitated adding a CMake FindBlosc.cmake file
* Add NCZarr support for the big-four filters provided by HDF5: shuffle, fletcher32, deflate (zlib), and szip
* Add a Codec defaulter (see docs/filters.md) for the big four filters.
* Make plugins work with windows by properly adding __declspec declaration.
### Misc. Non-Filter Changes
* Replace most uses of USE_NETCDF4 (deprecated) with USE_HDF5.
* Improve support for caching
* More fixes for path conversion code
* Fix misc. memory leaks
* Add new utility -- ncdump/ncpathcvt -- that does more or less the same thing as cygpath.
* Add a number of new test to test the non-filter fixes.
* Update the parsers
* Convert most instances of '#ifdef _MSC_VER' to '#ifdef _WIN32'
re: Issue https://github.com/Unidata/netcdf-c/issues/2096
The methods nc_set_var_chunk_cache_ints and nc_def_var_chunking_ints
are Fortran entry points for accessing the cache. They are not defined
if netcdf-c is built with --disable-hdf5.
Fix is to create dummy versions that do nothing and return NC_NOERR
when invoked. These dummy versions are defined when USE_HDF5 is false.
I think udata will be freed by the calback as soon as the callbacks are
set and H5Pclose is called, so the backward link should be set up at
that place.
Without this change, local_image_free may be called without a proper
reference to udata, which may raise an assertion:
assert(udata->fapl_image_ptr == udata->app_image_ptr);
while trying to open a broken file.
If a memory backed file is closed due to an aborted opening (i.e. a
broken file was attempted to be opened), udata may not be set. In this
case, the assertation checking for udata should not be triggered.
When opening an existing file for NC_WRITE access,
check whether the file was created originally with
attribute creation order tracking disabled and if
so, use that setting for subsequent attribute creation.
Also check the `mode` passed in to the open call
in case application is explicitly disabling the
attribute creation order tracking with the NC_NOATTRCREORD flag.
Work in progress / Proof of concept:
Add a capability to disable the tracking of attribute creation order.
See #2054 for details.
This PR adds a `NC_NOATTCREORD` define which can be passed int the
`mode` argument to `nc_create`. If it is present, then the
calls to set the attribute creation order tracking is disabled.
This should only be used for files in which you *know* that the
ordering of the attributes does not matter to *any* potential
readers of this database.
In HDF5 1.12.1, this was changed from optional to required per the changelog:
>H5Epush_ret() is a function-like macro that has been changed to
>contain a `do {} while(0)` loop. Consequently, a trailing semicolon
>is now required to end the `while` statement. Previously, a trailing
>semi would work, but was not mandatory.
This should be backward compatible with older version of HDF5.
re: https://github.com/Unidata/netcdf-c/issues/1996
Improve the error message and location that is reported when reading a filter with a variable that uses a filter that is not available on the reading platform.
This requires checking the availability of the filter, recording it, and failing when any attempt is made to read or write that variable. A test case was added for this in tst_filter.sh. Also, LOG level 0 message is generated giving the variable and the filter id.
Note that by design if there is no attempt to read or write the variable, then no error is reported; this means that, for example, ncdump -h will list the filter even though it is not actually available. This is important for allowing a user to see the filter details.
The netcdf-c code has to deal with a variety of platforms:
Windows, OSX, Linux, Cygwin, MSYS, etc. These platforms differ
significantly in the kind of file paths that they accept. So in
order to handle this, I have created a set of replacements for
the most common file system operations such as _open_ or _fopen_
or _access_ to manage the file path differences correctly.
A more limited version of this idea was already implemented via
the ncwinpath.h and dwinpath.c code. So this can be viewed as a
replacement for that code. And in path in many cases, the only
change that was required was to replace '#include <ncwinpath.h>'
with '#include <ncpathmgt.h>' and then replace file operation
calls with the NCxxx equivalent from ncpathmgr.h Note that
recently, the ncwinpath.h was renamed ncpathmgmt.h, so this pull
request should not require dealing with winpath.
The heart of the change is include/ncpathmgmt.h, which provides
alternate operations such as NCfopen or NCaccess and which properly
parse and rebuild path arguments to work for the platform on which
the code is executing. This mostly matters for Windows because of the
way that it uses backslash and drive letters, as compared to *nix*.
One important feature is that the user can do string manipulations
on a file path without having to worry too much about the platform
because the path management code will properly handle most mixed cases.
So one can for example concatenate a path suffix that uses forward
slashes to a Windows path and have it work correctly.
The conversion code is in libdispatch/dpathmgr.c, and the
important function there is NCpathcvt which does the proper
conversions to the local path format.
As a rule, most code should just replace their file operations with
the corresponding NCxxx ones defined in include/ncpathmgmt.h. These
NCxxx functions all call NCpathcvt on their path arguments before
executing the actual file operation.
In some rare cases, the client may need to directly use NCpathcvt,
but this should be avoided as much as possible. If there is a need
for supporting a new file operation not already in ncpathmgmt.h, then
use the code in dpathmgr.c as a template. Also please notify Unidata
so we can include it as a formal part or our supported operations.
Also, if you see an operation in the library that is not using the
NCxxx form, then please submit an issue so we can fix it.
Misc. Changes:
* Clean up the utf8 testing code; it is impossible to get some
tests to work under windows using shell scripts; the args do
not pass as utf8 but as some other encoding.
* Added an extra utf8 test case: test_unicode_path.sh
* Add a true test for HDF5 1.10.6 or later because as noted in
PR https://github.com/Unidata/netcdf-c/pull/1794,
HDF5 changed its Windows file path handling.
The XArray implementation that uses Zarr for storage
provides a mechanism to simulate named dimensions.
It does this by adding a per-variable attribute called
_ARRAY_DIMENSIONS. This attribute contains a list of names
to be matched against the shape values of the variable.
In effect a named dimension is created with the name
_ARRAY_DIMENSIONS(i) and length shape(i) for all i
in range 0..rank(variable).
Both read and write support is provided.
This XArray support is only invoked if the mode value
of "xarray" is defined. So for example, as in this URL.
````
https://s3.us-west-1.amazonaws.com/bucket/dataset#mode=nczarr,xarray,s3
````
Note that the "xarray" mode flag also implies mode flag "zarr", so the above
is equivalent to this URL.
````
https://s3.us-west-1.amazonaws.com/bucket/dataset#mode=nczarr,zarr,xarray,s3
````
The primary change to implement this was to unify the handling
of dimension references in libnczarr/zsync.
A test for this and other pure-zarr features was added as
nczarr_test/run_purezarr.sh
Other changes:
* Make sure distcheck leaves no files around.
* Change the special attribute flag DIMSCALEFLAG to HIDDENATTRFLAG
to support the xarray attribute.
* Annotate the zmap implementations with feature flags such as
WRITEONCE (for zip files).
than HDF5 1.8. This will allow for HDF5 features (VDS, SWMR, new references, etc...)
which may require to have a superblock greater than v2.
See for discussion Ref: Update HDF5 format compatibility Unidata#951
* Replace wholevar with more useful wholechunk optimization
* Add optimization to read multiple values at one time
* Replace NCDEFAULT_get/put_vars with native nczarr versions.
* Clarify chunk projection computations
* zdebdispatch.h
* Add more chunking test cases and re-enable run_chunkcases
* If !szip, then suppress deflate interference test
* Make H5Znoop(1) filter produce more information
* cleanup bzlib.c API
Primary Fixes:
* Add a whole variable optimization -- used in the rare case that nc_get/put_vara covers the whole of a variable and the variable has a single chunk.
* Fix chunking error when stride causes whole chunks to be skipped.
* Fix some memory leaks
* Add test cases
* Add one performance test to nczarr_test/. This uses the timer utils from unit_test: timer_utils.[ch].
* Move ncdumpchunks utility from ncdump to nczarr_test
Misc. Other Changes:
* Make check for aws libraries conditional on --enable-nczarr-s3
* Remove all but one bm tests from nczarr_test until they are working.
* Remove another dependency on HDF5 from supposedly non-HDF5 specific code; specifically hdf5_log_hdf5.
* Make the BAIL2 macro be hdf5 specific and replace elsewhere with an HDF5 independent equivalent.
* Move hdf5cache.c to libsrc4/nc4cache.c because it is used by nczarr.
* Modify unit_tests so that some of them are run even if using Windows.
* Misc. small bug fixes and refactors and memory leaks.
* Rename some conflicting tests for cmake.
* Attempted to make nc_perf work with cmake and failed.
The `http` field of the hdf5 info struct is not defined unless `ENABLE_HDF5_ROS3` or `ENABLE_BYTERANGE` or `ENABLE_S3_SDK` is defined. Based on a quick look at the code, I think that the `ENABLE_HDF5_ROS3` define is the relavant one here. Maybe a better fix is to check if any of them are defined...
Primary changes:
* Add an improved cache system to speed up performance.
* Fix NCZarr to properly handle scalar variables.
Misc. Related Changes:
* Added unit tests for extendible hash and for the generic cache.
* Add config parameter to set size of the NCZarr cache.
* Add initial performance tests but leave them unused.
* Add CRC64 support.
* Move location of ncdumpchunks utility from /ncgen to /ncdump.
* Refactor auth support.
Misc. Unrelated Changes:
* More cleanup of the S3 support
* Add support for S3 authentication in .rc files: HTTP.S3.ACCESSID and HTTP.S3.SECRETKEY.
* Remove the hashkey from the struct OBJHDR since it is never used.
If NCPROPERTIES_EXTRA (see configure.ac) is defined
but is null or empty, then an extra comma is being generated
at the end of _NCProperties global attribute.
Soln: check for null/empty NCPROPERTIES_EXTRA value.
There were some irregularities in the flags for handling NCZarr S3 support.
The primary change is to regularize the flags controlling this to the following.
1. Automake: --enable-nczarr-s3 and CMake: ENABLE_NCZARR_S3
2. Automake: --enable-nczarr-s3-tests and CMake: ENABLE_NCZARR_S3_TESTS
Flag 1 indicates that NCZarr should be built with S3 support enabled.
Flag 2 indicates that the NCZarr S3 tests should be run
These two flags are separate because running the NCZarr S3 tests
requires access to protected S3 resources. Currently, running
these tests is restricted to Unidata personnel. However, users
may want to enable S3 support even if they cannot run the tests.
It is, of course, an error to specify 2 without specifying 1.
Additionally, if the AWS S3 SDK library is not found, then the NCZARR S3
support and testing must be disabled. Otherwise an error is signaled
during the build.
Some of these NCZarr and S3 changes are propagated to nc-config.
Misc. Other Changes:
1. Allow testing for CYGWIN or MSVC in shell scripts.
2. Add specific test for HDF5 library version 1.10.6.
This is encoded as "HDF5_UTF8_PATHS" because that is the first
version where HDF5 properly supports it under Windows. This is used
in hdf5internal/nc4_ndf5_ansi_to_utf8.
3. Add a AM Conditional -- AX_IGNORE -- for use in testing
when it is desirable to temporarily suppress Makefile code.
4. Add MULTIFILTER flag to CMakeLists.txt
re: https://github.com/Unidata/netcdf-c/issues/1836
Revert the internal filter code to simplify it. From the user's
point of view, the only visible changes should be:
1. The functions that convert text to filter specs have had their signature reverted and have been moved to netcdf_aux.h
2. Some filter API functions now return NC_ENOFILTER when inquiry is made about some filter.
Internally,the dispatch table has been modified to get rid of the filter_actions
entry and associated complex structures. It has been replaced with
inq_var_filter_ids and inq_var_filter_info entries and the dispatch table
version has been bumped to 3. Corresponding NOOP and NOTNC4 functions
were added to libdispatch/dnotnc4.c. Also, the filter_action entries
in dispatch tables were replaced for all dispatch code bases (HDF5, DAP2,
etc). This should only impact UDF users.
In the process, it became clear that the form of the filters
field in NC_VAR_INFO_T was format dependent, so I converted it to
be of type void* and pushed its management into the various dispatch
code bases. Specifically libhdf5 and libnczarr now manage the filters
field in their own way.
The auxilliary functions for parsing textual filter specifications
were moved to netcdf_aux.h and were renamed to the following:
* ncaux_h5filterspec_parse
* ncaux_h5filterspec_parselist
* ncaux_h5filterspec_free
* ncaux_h5filter_fix8
Misc. Other Changes:
1. Document NUG/filters.md updated to reflect the changes above.
2. All the old data types (structs and enums)
used by filter_actions actions were deleted.
The exception is the NC_H5_Filterspec because it is needed
by ncaux_h5filterspec_parselist.
3. Clientside filters were removed -- another enhancement
for which no-one ever asked.
4. The ability to remove filters was itself removed.
5. Some functionality needed by nczarr was moved from libhdf5
to libsrc4 e.g. nc4_find_default_chunksizes
6. All the filterx code was removed
7. ncfilter.h and nc4filter.c no longer used
Misc. Unrelated Changes:
1. The nczarr_test makefile clean was leaving some directories; so
add clean-local to take care of them.
re: Issue https://github.com/Unidata/netcdf-c/issues/1848
The existing Virtual File Driver built to support byte-range
read-only file access is quite old. It turns out to be extremely
slow (reason unknown at the moment).
Starting with HDF5 1.10.6, the HDF5 library has its own version
of such a file driver. The HDF5 developers have better knowledge
about building such a driver and what incantations are needed to
get good performance.
This PR modifies the byte-range code in hdf5open.c so
that if the HDF5 file driver is available, then it is used
in preference to the one written by the Netcdf group.
Misc. Other Changes:
1. Moved all of nc4print code to ncdump to keep appveyor quiet.
Older HDF5 libraries do not support virtual datasets but could otherwise
be supported by netCDF4. This change removes the special case to handle
HDF5 virtual datasets if the installed HDF5 version does not support
virtual datasets.
re: Github Issue https://github.com/Unidata/netcdf-c/issues/1826
It turns out that the common get code (NC4_get_vars) in libhdf5
(and libnczarr) has an optimization where it does not attempt to
read from the file if the file is all fill values. Rather it
just fills the output buffer with the fill value. The problem
is that -- in that case -- it forgets that conversion might still be
needed. So the conversion never occurs and the raw bits of
the fill data are stored directly into the memory space.
Solution: move some code around to properly do the
conversion no matter how the data was obtained.
Added a test cases nc_test4/test_fillonly.sh and
nczarr_test/test_fillonlyz.sh
It seems like it is part of the design of HDF5 virtual datasets that
objects within a file remain opened while the files is aready "closed".
Setting the fclose degree to SEMI would cause the library to bail out.
This commit makes nc_test4/tst_virtual_dataset succeed.
See also Unidata/netcdf-c#1799
In case HDF5 adds more storage specifications, netcdf4 should be able to
cope with them by default. Further specializations could be added
nonetheless.
disengagement of enable-netcdf4 from enable-hdf5.
That is, with the advent of nczarr, it is possible
to turn off hdf5 but still need netcdf-4 enabled
because nczarr uses libsrc4, but not libhdf5.
This change involves a bunch of things:
1. Modify configure.ac and CMakelist to make enable_hdf5
control if hdf5 support is provided. For back compatibility,
disable-netcdf4 is treated as disable-hdf5. But internally,
netcdf4 support is controlled only by the enabling of formats
that require it.
2. In support of #1, modify .travis.yml to use enable/disable-hdf5
instead of enable/disable-netcdf4.
3. test_common.in is modified to track selected features,
including enable-hdf5 and enable-s3-tests. This is used in
selected tests that mix netcdf-3 and netcdf4 tests.
4. The conflation of USE_HDF5 and USE_NETCDF4 is common in
code, tests, and build files, so all of those had to be weeded out.
5. It turns out that some of the NC4_dim functions really are HDF5 specific,
but are not treated as such. So they are moved from nc4dim.c to
hdf5dim.c or hdf5dispatch.c
6. Some generic functions in libhdf5 can be (and were) moved to libsrc4.
cloud using a variant of the Zarr protocol and storage
format. This enhancement is generically referred to as "NCZarr".
The data model supported by NCZarr is netcdf-4 minus the user-defined
types and the String type. In this sense it is similar to the CDF-5
data model.
More detailed information about enabling and using NCZarr is
described in the document NUG/nczarr.md and in a
[Unidata Developer's blog entry](https://www.unidata.ucar.edu/blogs/developer/en/entry/overview-of-zarr-support-in).
WARNING: this code has had limited testing, so do use this version
for production work. Also, performance improvements are ongoing.
Note especially the following platform matrix of successful tests:
Platform | Build System | S3 support
------------------------------------
Linux+gcc | Automake | yes
Linux+gcc | CMake | yes
Visual Studio | CMake | no
Additionally, and as a consequence of the addition of NCZarr,
major changes have been made to the Filter API. NOTE: NCZarr
does not yet support filters, but these changes are enablers for
that support in the future. Note that it is possible
(probable?) that there will be some accidental reversions if the
changes here did not correctly mimic the existing filter testing.
In any case, previously filter ids and parameters were of type
unsigned int. In order to support the more general zarr filter
model, this was all converted to char*. The old HDF5-specific,
unsigned int operations are still supported but they are
wrappers around the new, char* based nc_filterx_XXX functions.
This entailed at least the following changes:
1. Added the files libdispatch/dfilterx.c and include/ncfilter.h
2. Some filterx utilities have been moved to libdispatch/daux.c
3. A new entry, "filter_actions" was added to the NCDispatch table
and the version bumped.
4. An overly complex set of structs was created to support funnelling
all of the filterx operations thru a single dispatch
"filter_actions" entry.
5. Move common code to from libhdf5 to libsrc4 so that it is accessible
to nczarr.
Changes directly related to Zarr:
1. Modified CMakeList.txt and configure.ac to support both C and C++
-- this is in support of S3 support via the awd-sdk libraries.
2. Define a size64_t type to support nczarr.
3. More reworking of libdispatch/dinfermodel.c to
support zarr and to regularize the structure of the fragments
section of a URL.
Changes not directly related to Zarr:
1. Make client-side filter registration be conditional, with default off.
2. Hack include/nc4internal.h to make some flags added by Ed be unique:
e.g. NC_CREAT, NC_INDEF, etc.
3. cleanup include/nchttp.h and libdispatch/dhttp.c.
4. Misc. changes to support compiling under Visual Studio including:
* Better testing under windows for dirent.h and opendir and closedir.
5. Misc. changes to the oc2 code to support various libcurl CURLOPT flags
and to centralize error reporting.
6. By default, suppress the vlen tests that have unfixed memory leaks; add option to enable them.
7. Make part of the nc_test/test_byterange.sh test be contingent on remotetest.unidata.ucar.edu being accessible.
Changes Left TO-DO:
1. fix provenance code, it is too HDF5 specific.
It is possible for the values stored to `file_value_size` to overrun the storage capacity of a 32-bit integer. The value does need to store negative values potentially, so can be `size_t` or `hsize_t`, so use the `hssize_t` which is a signed 64-bit value. Could also use `ssize_t`, but that is not used in this routine...
re: Github issue https://github.com/Unidata/netcdf-c/issues/1713
If nc_def_var_filter or nc_def_var_deflate or nc_def_var_szip is
called multiple times with the same filter id, but possibly with
different sets of parameters, then the first invocation is
sticky and later invocations are ignored. The desired behavior
is to have the last invocation be used.
This PR implements that desired behavior, with some special
cases. If you call nc_def_var_deflate multiple times, then the
last invocation rule applies with respect to deflate. However,
the shuffle filter, if enabled, is always applied just before
applying deflate.
Misc unrelated changes:
1. Make client-side filters be disabled by default
2. Fix the definition of uintptr_t and use in oc2 and libdap4
3. Add some test cases
4. modify filter order tests to use plugin filters rather
than client-side filters
The current library seems to have some behavior which is N^2 in the number of vars in a file.
The `NC4_inq_dim` routine calls down to `nc4_find_dim_len` which iterates through each `var` in the file/group and calls `find_var_dim_max_length` on each var and finds the largest length of the dim on each of those vars. This is done only for unlimited vars.
I have a file with 129 dim and 1630 vars. The unlimited dimension is of length 41. In my test program, I am reading data from 4 files which have the same dim and var count and reading every 4th time step (unlimited dimension). If I run a profile, I see that 98.2% of the program time is in the `nc_get_vara_float` call tree and most of that is in `find_var_dim_max_length` (94.8%).
There are 66,142 calls to `nc_get_vara_float` resulting in 107,307,290 calls to `find_var_dim_max_length` with twice that number of calls to `malloc/free` and calls to 5 HDF5 routines. All of this, at least in my case, to return the same `41` each time.
The proof of concept patch here will check whether the file is read-only (or no_write) and if so, it will cache the value of the dim length the first time it is calculated. With this change, my example run is sped up by a factor of 60. The time for `NC4_inq_dim` and below drops from 97.2% down to 2.7%.
I'm not sure whether this is the correct fix, or if there is some behavior that I am overlooking, but my users would definitely like a 10 second run compared to a 10 minute run...
This is on current Netcdf master branch.
I will try to attach some valgrind/callgrind profiles.
nc4internal.c contains code to free the format_XXX_info
fields. Since these are format specific, this code
was moved to the dispatch code (libhdf5 and libhdf4
in the current case).
Additionally, there are some fields in nc4internal.h (e.g.
dimscale fields) that are specific to HDF5 and have been moved
to the corresponding HDF5 data structures and code.
Misc. other changes:
1. NC_VAR_INFO_T->hdf5_name renamed to alt_name to avoid
implying it is necessarily HDF5 specific.
2. prefix NC_FILE_INFO_T with an instance of NC_OBJ for consistency.
this also requires wrapping move_in_NCList() to keep
hdr.id consistent.
