Forum Discussion
Advocate IIICan anyone explain why this works?
I see your point. I was similarly bewildered when first presented with this type of formulae.
There are two tricky subjects here, by order of importance:
1. What 'Date'[Full Date] actually refers to.
2. The nuances of how ALL( ) actually works
The behaviour of ALL depends on where it is used. It should probably have two different names. Since you're already familiar with the Italian gurus, check out this article for details. ALL can work either by removing filters or by ignoring filters. It sounds like the same but it sure ain't.
Let us walk through your example:
The filter context
We have 'Date'[Full Date] in the rows of our pivot table. That determines the filter context.
Inside FILTER( )
FILTER( ) first computes the table it will operate on. In this case, ALL('Date'). That is the full 'Date' table in your data model. Let's call it Table=ALL('Date').
Nota bene: ALL( ) here ignores filter context but it does NOT remove it, so filter context will be effective everywhere else.
Table has a 'Date'[Full Date] column as well, an instance different from the one in the pivot that we marked in red above. We'll use purple for Table's : 'Date'[Full Date]. So we have one name referring to two different instances. We'll have to see how we tell them apart.
FILTER( ) and filter context interaction
Once it has Table, FILTER( ) starts scanning it to check whether each row complies with the condition:
'Date'[FullDate] < MIN ( 'Date'[FullDate] )
'Date'[FullDate] is the value in the current row of Table. MIN('Date'[FullDate]), however, is the minimum of the values of 'Date'[Full Date] coming from the pivot table (filter context). This is the crux.
Conclusions
You do have a valid point in that MIN('Date'[FullDate]) could actually be referring to MIN('Date'[FullDate]), in which case every row would certainly be filtered out.
How does DAX discriminate between 'Date'[FullDate] and 'Date'[FullDate]?
Basically, when you use the "naked" column, i.e. the column on its own, it's 'Date'[FullDate]. In any other case you'll be referring to 'Date'[Full Date]. It's been built that way so that you can refer to both instances and build powerful code like the one you've shown.
As a final point, bear in mind that filter context affects both arguments in
FILTER(<table>; <filter expression>)
In this case the filter context is ignored in the first argument because we are using ALL( ). It'd be another story if we were using VALUES( ).
Does that help? (I do hope so, as it took some time to put together :smileywink:)
Advocate IIIHi AIB,
Here is my logic (details in another reply):
Since ALL() removes the filter context, both 'Date'[FullDate] should be under the same filter context (which is, no filter context at all), that's the reason I think it should return 0. After all, none of the element should be smaller than the minimum.
Community ChampionI see your point. I was similarly bewildered when first presented with this type of formulae.
There are two tricky subjects here, by order of importance:
1. What 'Date'[Full Date] actually refers to.
2. The nuances of how ALL( ) actually works
The behaviour of ALL depends on where it is used. It should probably have two different names. Since you're already familiar with the Italian gurus, check out this article for details. ALL can work either by removing filters or by ignoring filters. It sounds like the same but it sure ain't.
Let us walk through your example:
The filter context
We have 'Date'[Full Date] in the rows of our pivot table. That determines the filter context.
Inside FILTER( )
FILTER( ) first computes the table it will operate on. In this case, ALL('Date'). That is the full 'Date' table in your data model. Let's call it Table=ALL('Date').
Nota bene: ALL( ) here ignores filter context but it does NOT remove it, so filter context will be effective everywhere else.
Table has a 'Date'[Full Date] column as well, an instance different from the one in the pivot that we marked in red above. We'll use purple for Table's : 'Date'[Full Date]. So we have one name referring to two different instances. We'll have to see how we tell them apart.
FILTER( ) and filter context interaction
Once it has Table, FILTER( ) starts scanning it to check whether each row complies with the condition:
'Date'[FullDate] < MIN ( 'Date'[FullDate] )
'Date'[FullDate] is the value in the current row of Table. MIN('Date'[FullDate]), however, is the minimum of the values of 'Date'[Full Date] coming from the pivot table (filter context). This is the crux.
Conclusions
You do have a valid point in that MIN('Date'[FullDate]) could actually be referring to MIN('Date'[FullDate]), in which case every row would certainly be filtered out.
How does DAX discriminate between 'Date'[FullDate] and 'Date'[FullDate]?
Basically, when you use the "naked" column, i.e. the column on its own, it's 'Date'[FullDate]. In any other case you'll be referring to 'Date'[Full Date]. It's been built that way so that you can refer to both instances and build powerful code like the one you've shown.
As a final point, bear in mind that filter context affects both arguments in
FILTER(<table>; <filter expression>)
In this case the filter context is ignored in the first argument because we are using ALL( ). It'd be another story if we were using VALUES( ).
Does that help? (I do hope so, as it took some time to put together :smileywink:)
- markus_zhang
Thanks AIB, this makes everything clear.
"Basically, when you use the "naked" column, i.e. the column on its own, it's 'Date'[FullDate]. In any other case you'll be referring to 'Date'[Full Date]."
This is the key! Does that mean, whenever I use a function like MIN(), MAX(), FIRSTNONBLANK(), etc. to wrap 'Date'[FullDate], it ignores the ALL() effect?
- AlB
Kind of. The way you describe it is, being strict, innacurate but I think you get the point.
The ALL( ) here provides the full 'Date' table as table to be scanned in FILTER(). Its effect finishes there.
The second argument in FILTER() is also affected by filter context but when you use the naked column 'Date'[FullDate] you are referring to 'Date'[FullDate], the column in the table being scanned, which we just said has all rows of 'Date'. If you wrap 'Date'[FullDate] in whatever, MIN( ), MAX( ), etc. the filter context applies because you are actually referring to 'Date'[FullDate].
Now think what would happen if instead of ALL( ) we had VALUES( ) as below. That will help you nail it.
FILTER ( VALUES( 'Date' ), 'Date'[FullDate] < MIN ( 'Date'[FullDate] ) )- markus_zhang
If we use VALUES() instead of ALL() then the current filter context should still be effective, so I guess it's completely different story. Thanks for the help!