You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The LDAS folks were finding that turning collections to instantaneous from time-averaged in History was changing the GEOSldas.x checkpoint. I looked into it and it was indeed the case. The issue though is the bit shaving in the binary code path.
Looks like since v2.8.6 of MAPL binary output has had a bit-shaving option. Turns out there's a possible bug. When bit shaving basically the "output state" of the collection is passed to an optional routine to bit shave based on user input.
Looks like if instantaneous output, we are bit-shaving the actual fields from the component. Because of the friendliness business in MAPL, many exports are more than just diagnostic and this can change the model state. I confirmed that turning off bit shaving got rid of the non-zero diff.
If it is time-averaged, we've made a new buffer for the accumulation and this is what is bit shaved. When the accumulation is done, we shave that buffer, it has no effect on the component fields.
So if we are bit shaving instataneous output, we need to make a copy.
The text was updated successfully, but these errors were encountered:
The LDAS folks were finding that turning collections to instantaneous from time-averaged in History was changing the GEOSldas.x checkpoint. I looked into it and it was indeed the case. The issue though is the bit shaving in the binary code path.
Looks like since v2.8.6 of MAPL binary output has had a bit-shaving option. Turns out there's a possible bug. When bit shaving basically the "output state" of the collection is passed to an optional routine to bit shave based on user input.
Looks like if instantaneous output, we are bit-shaving the actual fields from the component. Because of the friendliness business in MAPL, many exports are more than just diagnostic and this can change the model state. I confirmed that turning off bit shaving got rid of the non-zero diff.
If it is time-averaged, we've made a new buffer for the accumulation and this is what is bit shaved. When the accumulation is done, we shave that buffer, it has no effect on the component fields.
So if we are bit shaving instataneous output, we need to make a copy.
The text was updated successfully, but these errors were encountered: