search

ratt-ru / QuartiCal

CubiCal, but with greater power.
MIT License
8 stars 4 forks source link

segfault on single correlation MS #100

Closed landmanbester closed 3 years ago

landmanbester commented 3 years ago

I ran into this while trying to calibrate a tiny MS I split out for testing purposes. Definitely an edge case. MS and parset are on oates at /home/bester/projects/ESO137/msdir/ms1_primary_scan1_corrXX_chan950to1050.ms and /home/bester/projects/ESO137/qcal_1gc.yaml respectively. I can reproduce the segfault locally as well. Interestingly two consecutive runs produces slightly different errors viz.

2021-08-12 08:42:45 | SUCCESS | executor:_execute | 3.97 seconds taken to build graph.
[#######################                 ] | 59% Completed | 38.4sSegmentation fault (core dumped)

and

[#######################                 ] | 59% Completed |  1.0sfree(): invalid next size (normal)
Aborted (core dumped)

Any idea what could be going wrong here @JSKenyon?

JSKenyon commented 3 years ago

As discussed, the single correlation path is still a work in progress. For now, setting input_ms.select_corr=[0,0] will solve the problem (by virtue of making the problem appear diagonal). We can leave this open as a reminder to me to communicate when this changes.

landmanbester commented 3 years ago

Setting input_ms.select_corr=[0,0] solves the problem

JSKenyon commented 3 years ago

The master branch should now work correctly for single correlation MSs, provided you use the complex or slow_complex solvers. I am still working on the parameterised terms.

JSKenyon commented 3 years ago

I am going to close this - I believe that all the solvers now correctly handle 1,2 and 4 correlation data. Please reopen if you encounter problems.