From: Francesc A. <fa...@py...> - 2004-10-01 12:40:18
|
Hi, I'm in the final phase of testing PyTables 0.9 before releasing it. As the number of supported platforms is increasing, I'm asking for your help here. As of now, I've successfully tested PyTables mainly with Linux machines, both 32-bits (Pentium IV) and 64-bits (Intel Itanium, AMD Opteron). However, I would like to test, at very least, the Windows and MacOSX implementation. If someone is able to test a BSD, Solaris or Irix platforms, much better :) The procedure of compilation/installation has not been updated in documentation yet, but it should be fairly similar to 0.8.1. You should be able to run PyTables with both HDF5 1.6.2 and HDF5 1.6.3-post1, and either numarray 1.0 or numarray 1.1 (although I've recently only tested numarray 1.1). You can find the 0.9 pre-release in: http://pytables.sourceforge.net/html/GetIt.html and clicking in "Preview of the forthcoming 0.9 release" Please, have in mind that the new "official" version of HDF5, namely 1.6.3, version came out with a nasty bug that prevents to delete an Index object (which has been introduced in PyTables 0.9) and besides, it causes a segmentation fault if you try to do that. I've asked to the NCSA to provide binary libraries with that bug corrected (the -post1 version), and they are willing to do that, at least for Windows and MacOSX platforms. So, any user on Win or MacOSX can test HDF5 1.6.3-post1 binaries as well? I'm attaching below the message from Elena Pourmal (the Maintenance and Support Team Leader) with the location for these libraries and asking for confirmation that the libraries works well. One thing more, a Windows user has already told me that he is having problems with the newly introduced calls to nextafter and nextafterf mathematical functions that appear not to be on "math.h", which is the official place in C99 standard. Someone can provide a patch for this (using the Microsoft C Library)?. Many thanks for your time! ---------- Missatge transmes ---------- Subject: binaries with applied patch Re: Segmentation fault with 1.6.3 Date: Divendres 01 Octubre 2004 01:22 From: Elena Pourmal <epo...@nc...> To: Francesc Alted <fa...@py...> Cc: hd...@nc... Francesc, I've built binaries with applied path for Windows, Linux 2.4 and Mac OSX. Binaries are not configured with szip support but gzip support is included. Files are on our ftp server in my outgoing directory: ftp://hdf.ncsa.uiuc.edu/pub/outgoing/epourmal/patch Could you please test at least one binary distribution and make sure it works for you? If it works, we will move the files to our official distribution site and we will let you know the exact location. If you want to build from scratch yourself, just drop H5Distore.c file into the src directory of the hdf5-1.6.3 source. Thank you! Elena At 01:33 PM 9/28/2004, you wrote: >A Dimarts 28 Setembre 2004 20:16, vareu escriure: > > Francesc, > > > > We do not keep old binaries on the ftp server. Do you know for which > > platforms your users need HDF5 binaries? > > > >I was thinking mostly in Windows platforms, but I know that there are some >other PyTables users using MacOSX. > >Another unfortunate consequence of the bug in the official release is that, >normally, this is the one choosed by package maintainers for including in >linux distributions (at least, this is the case for Debian), so most >probably, most Linux distributions will distribute the buggy release in form >of binary packages. > >Anyway, if it is difficult to you to integrate the patch for 1.6.3, I'm >afraid that I'll have to disable indexing features for the forthcoming >pytables 0.9. That is not necessarily too bad, only that users will have to >wait for this functionality until HDF5 1.6.4 or whatever. > >Regards, > >-- >Francesc Alted ----------------------------------------------------------------------------------------------- Elena Pourmal, HDF QA, Maintenance and Support Team Leader NCSA University of Illinois at Urbana-Champaign 605 E. Springfield Ave. Champaign, IL 61820 epo...@nc... (217)333-0238 (office) (217)244-1987 (fax) ---------------------------------------------------------------------------------------------- ------------------------------------------------------- -- Francesc Alted |