Re: [Pytables-users] PyTables 1.1-beta3

[Francesc A. <fa...@ca...>]

Ashley,

On Thursday 23 June 2005 04:44, you wrote:
> I tested the 20050622 snapshot.  Here are few changes I made.
>
> AttributeSet now preloads all the attributes.  I've been pickling
> classes in the attributes that may not available when the file is
> opened (its a bad design --  I'll change it to use class names one
> day), and this raises an exception in AttributeSet.__init__.  My
> solution was to add ImportError to the "allowable exceptions" in
> AttributeSet.__getattr__.  I did notice that the documentation for
> pickling says that unpickling may raise errors other than
> UnpicklingError.
>
>
> --- pytables/tables/AttributeSet.py     2005-06-16 12:43:41.000000000
> +1000
> +++ pytables22/tables/AttributeSet.py   2005-06-23 12:20:33.000000000
> +1000
> @@ -205,7 +205,7 @@
>           if type(value) is str and value and value[-1] == ".":
>               try:
>                   retval = cPickle.loads(value)
> -            except cPickle.UnpicklingError:
> +            except (cPickle.UnpicklingError, ImportError):
>                   retval = value
>           else:
>               retval = value

Yup, you suggestion has been added. Perhaps something like:

     try:
     
     except:
     
would be better? Ivan, do you see something wrong with this?

> The other two are just simple mistakes:
>
>
> --- pytables/tables/File.py     2005-06-18 12:36:31.000000000 +1000
> +++ pytables22/tables/File.py   2005-06-23 12:23:16.000000000 +1000
> @@ -912,7 +912,7 @@
>           """
>           srcObject = self.getNode(where, name=name)
>           dstObject = self.getNode(dstnode)
> -        object._v_attrs._f_copy(dstnode)
> +        srcObject._v_attrs._f_copy(dstnode)
>       def copyChildren(self, srcgroup, dstgroup,

Ops. Solved and added a couple of tests to check File.copyNodeAttrs()
just in case.

>
> --- pytables/tables/Table.py    2005-06-21 13:25:17.000000000 +1000
> +++ pytables22/tables/Table.py  2005-06-23 12:21:52.000000000 +1000
> @@ -1666,9 +1666,9 @@
>                       index = self._v_file.getNode(indexpathname)
>                       # Get the filters values
>                       self.indexprops.filters = self._g_getFilters()
> -                    description[colname].indexed = 1
> +                    getattr(description, colname).indexed = 1
>                   except NodeError:
> -                    description[colname].indexed = 0
> +                    getattr(description, colname).indexed = 0
>           # Add a possible IndexProps property to that
>           if hasattr(self, "indexprops"):

Indeed, you are right. Fixed!

> The tests pass with the exception of a segmentation fault that occurs
> intermittently.
>
> The platform is:
>
> -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> =-=-=-=
> PyTables version:  1.1-beta4
> Extension version: $Id: utilsExtension.pyx 1019 2005-06-20 12:54:50Z
> faltet $
> HDF5 version:      1.6.4
> numarray version:  1.4.0
> Zlib version:      1.2.2
> LZO version:       1.08 (Jul 12 2002)
> UCL version:       1.03 (Jul 20 2004)
> BZIP2 version:     1.0.2 (30-Dec-2001)
> Python version:    2.3.5 (#1, Mar 20 2005, 20:38:20)
> [GCC 3.3 20030304 (Apple Computer, Inc. build 1809)]
> Platform:          darwin-Power Macintosh
> Byte-ordering:     big
> -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> =-=-=-=
>
> I get occasional failures with from ``test_earray`` when run inside
> ``test_all``::
>
>      Checking enlargeable array iterator ... Segmentation fault
>
> Its not a reproducible error.  If I run only ``test_earray`` with
> ``heavy=1``, it succeeds::
[...]
> The crash log is::
>
>      Exception:  EXC_BAD_ACCESS (0x0001)
>      Codes:      KERN_INVALID_ADDRESS (0x0001) at 0x024f8000
>
>      Thread 0 Crashed:
>      0   utilsExtension.so     0x016b4ae4 ucl_nrv2e_decompress_safe_8
> + 684 (n2e_d.c:122)
>      1   utilsExtension.so     0x016af480 ucl_deflate + 320 (H5Zucl.c:

Mmm, and it is always this UCL call which crash? Perhaps this might a
consequence of a bug in UCL for PowerPC platforms, o perhaps (more
possible) a bad implementation of UCL support on my part. I've
slightly modified the size for buffer compression: can you give it a
try so as to see if this symptom has bettered?

If problems with UCL compression library remains for PowerPC, I'll
have to disable support for it, at least until I get a fix :-/

> Hope this helps.

Of course it does! PowerPC was not tested yet for PyTables 1.1.
Thanks!

-- 
Francesc