populate_mosaic_basic overwritten in step

[braingram]

The metadata blending performed by populate_mosaic_basic:

romancal/romancal/resample/resample.py

Line 875 in 826a972

def populate_mosaic_basic(

is later overwritten by:

• update_exposure_times:

  romancal/romancal/resample/resample.py

  Line 533 in 826a972

  def update_exposure_times(self, output_model, exptime_tot):

• l2_into_l3_meta:

  romancal/romancal/resample/resample.py

  Line 734 in 826a972

  def l2_into_l3_meta(l3_meta, l2_meta):

The test PR #1597 adds a test that runs the resample step on 2 inputs with different metadata:

                {
                    "meta.observation.visit": 1,
                    "meta.observation.pass": 1,
                    "meta.observation.segment": 1,
                    "meta.observation.program": 1,
                    "meta.instrument.optical_element": "F158",
                    "meta.instrument.name": "WFI",
                },
                {
                    "meta.observation.visit": 2,
                    "meta.observation.pass": 2,
                    "meta.observation.segment": 2,
                    "meta.observation.program": 2,
                    "meta.instrument.optical_element": "F062",
                    "meta.instrument.name": "WFI",
                },
            ),

The mosaic model contains:

            {
                "visit": 1,
                "pass": 1,
                "segment": 1,
                "optical_element": "F158",
                "instrument": "WFI",
            },

This shows that the metadata assigned by populate_mosaic_basic is being overwritten since it would set:

• visit: -1
• pass: -1
• segment: -1

[schlafly]

Thanks. Looking at the code here---the update_exposure_times bit is redundant but at least the calculations look the same. Meanwhile l2_into_l3_meta "contributes" only coordinates and so should probably get deleted with the "coordinates" contribution folded into populate_mosaic_basic; the other bits it does are inferior to what's happening in populate_mosaic_basic.

Anyway, I agree that overwriting without the blending is wrong and the redundant calculations are problematic---just making sure we agree on the scope of the bug here. Thanks!

[nden]

@stscijgbot-rstdms

[stscijgbot-rstdms]

This issue is tracked on JIRA as RCAL-1006.

[braingram]

Thanks. Looking at the code here---the update_exposure_times bit is redundant but at least the calculations look the same. Meanwhile l2_into_l3_meta "contributes" only coordinates and so should probably get deleted with the "coordinates" contribution folded into populate_mosaic_basic; the other bits it does are inferior to what's happening in populate_mosaic_basic.

Thanks for taking a look. The calculations are different currently. To use meta.observation.visit as an example let's say there are 2 inputs with visits 42 and 26.

populate_mosaic_basic would set the output meta.observation.visit==-1 since 42 and 26 are different. However this would then be overwritten by l2_into_l3_meta which would set the output meta.observation.visit==42 (setting it to the visit of the first input).

So with the current code on main the above example would output a model with meta.observation.visit==42.

[schlafly]

Sorry, I was referring specifically to update_exposure_times when I said that the calculations are redundant but the same.

I agree that l2_into_l3 should get deleted, moving only its "coordinates" copying into populate_mosaic_basic.

[braingram]

Thanks! Good to know. I will keep that in mind for the changes needed to use the stcal resample.