Clarify precedence (cloud > shadow > snow > water)

[benjimin]

Expressing this as unconvolutedly as possible. (Should also perform slightly faster, and use slightly less memory.)

[neilflood]

I don't understand what problem you are trying to solve. The original code is pretty clear.

The speed is dominated by I/O, and also by the shadow search phase (depending on size and number of clouds). The memory usage is already very efficient, due to the use of RIOS for raster processing.

Could you elaborate on your concerns?

[benjimin]

My concern is readability. I sometimes work with data that belongs to multiple categories (such as shadow on water) and debug suspected misclassifications, and I've occasionally needed to confirm the category precedence order. More than once in the past, I've looked up this section of the code to remind myself or verify which category gets reported if multiple apply. I found the original version to be convoluted, but unnecessarily so:

• The function has 40% more code (e.g., 7 extra raster operations are performed).
• Several more objects get modified between uses, instead of being left constant. (I find it more difficult to correctly reason through steps that involve complex state.)
• More names are introduced, including one that already has an exact synonym.
• One array is initialised twice (to two different values).
• The code is organised confusingly (i.e. at first glance it reads as though snow overrides cloud, although the reverse is performed).

I propose this very minor change:

• Maintaining identical behaviour.
• Improved readability (where it documents something critical).
• Performance is not impacted (and is actually improved: fewer computations and 3 fewer 8bit rasters held simultaneously in memory. BTW even if each individual process is limited by I/O latency, still CPU and memory efficiency can be relevant to big data parallel processing, such as this package is used for.)

[neilflood]

OK, fair enough. I have tested this using LC08_L1TP_091085_20180813_20180828_01_T1, which has some of all classes present, and it gives identical results.