Custom Query (116 matches)

As far as I can tell the automatic detection of buffer size is causing two issues (closely related):

• XIOS will raise an error if a context has no enabled output file which it has no good reasons to do.
• XIOS can fail if internal messages are bigger than the estimated output data.

This can probably be solved by always allocating a buffer with a fixed minimum size we know will always be sufficient to allocate the internal messages.

It seems the recent grid related changes have caused a regression which affects the zoom feature. If some processes have nothing to output, XIOS will deadlock. This can be reproduced using the "test_complete" example after r565.

This is possibly caused by the "pseudo-global" communication style used by XIOS, if one client skips the communication with the server, a deadlock will occur.

XIOS crashes when outputting a field defined on a 2D grid made of 2 distributed axis.

Further investigations show that the indexes used to extract the data received from the clients are completely wrong which causes a segmentation fault.

I am attaching a simple test case which can be used to reproduce the problem in attached mode. The code is pretty limited so it must be run on N processes with N a perfect square (for example N = 4 or N = 9).