Inherited Tag Values are not Overwritten [💰$20 for 100% human contributions]

[nobodyinperson]

Thank you very much for hledger, it is awesome! 🎉

Apparently there is a misleading behaviour when inheriting values of tags.

The docs (https://hledger.org/1.27/hledger.html#tags-1) mention that tags are inherited, but only mention value-less tags.

Consider this book.hledger:

2022-11-17 Aldi
    ; concerns: me
    Assets               -30 €
    Costs:Food            20 €
    Loaned                10 €  ; concerns: you, note: 👈 This should OVERWRITE the concerns: me from above, shouldn't it?

For consistency, higher-level tag values should overwrite tags values lower down the chain, right? At least a transaction/account shouldn't have two values of the same tag at once, right? But:

❯ hledger --version
hledger 1.27.1, linux-x86_64

❯ hledger -f book.hledger balance  --pivot=concerns
               -10 €  me
                10 €  you
--------------------
                   0  
# ☝️ This is expected.


❯ hledger -f book.hledger register Loaned  --pivot=concerns
2022-11-17 Aldi    you              10 €          10 €
# ☝️ This is expected.

❯ hledger -f book.hledger register tag:concerns=me --pivot=concerns
2022-11-17 Aldi    me              -30 €         -30 €
                   me               20 €         -10 €
                   you              10 €             0 # 👈 why is this line here? I explicitly filtered for tag:concerns=me but a line with concerns=you appears?

❯ hledger -f book.hledger register Loaned tag:concerns=me
2022-11-17 Aldi          Loaned                 10 €          10 €
# ☝️ Why does this appear? 

❯ hledger -f book.hledger register Loaned tag:concerns=you
2022-11-17 Aldi          Loaned                 10 €          10 €
# ☝️ This is expected.

Conclusion

It seems that tag values are not overwritten down the chain but somehow still kept around. This can lead to confusion and inconsistencies in register for example.

Proposal

I propose that tag values down the line should overwrite higher-level tag values.

[chrislemaire]

Looks like it should be easy enough to change by tinkering with the postingAllTags function in Posting.hs. My solution would be to have this function filter the tags coming from transactions to only include those that have not been used by the posting already. This way, we let postings override all tags of the same name of their parents. My remaining question is whether this is desired behaviour, which a dev might be able to answer better.

Also, would it be enough to do this just to postings, or might it be useful to also do this for accounts inheriting tags from parent accounts?

[simonmichael]

Thanks for the discussion and PR, and sorry for the slow response. After looking into it.. yes, though docs don't explain it well, tags are multi-valued; they collect and add new values, rather than overriding a single value. This gets tricky, there are multiple cases to consider:

1. Same tag on sibling items, eg the same tag on multiple transactions
2. Inheriting tags from a parent item, eg posting inheriting tags from (a) transaction or (b) account
3. Acquiring tags from a child item, eg transaction acquiring tags from postings

Currently in all of these cases we add rather than override values. This explains the behaviour you're seeing, and has the virtue of being simple and consistent.

You expect it to override instead, in cases 2a and 2b ? This sounds reasonable, if it does not add confusion. I feel there might be ambiguities lurking, we should work out a complete set of tests.

[simonmichael]

PS re this example:

❯ hledger -f book.hledger register tag:concerns=me --pivot=concerns
2022-11-17 Aldi    me              -30 €         -30 €
                   me               20 €         -10 €
                   you              10 €             0 # 👈 why is this line here? 

I explicitly filtered for tag:concerns=me but a line with concerns=you appears?

The Loaned posting also inherits me from the transaction, though you is shown as it is the most specific value for the posting (a partial kind of overriding). To exclude the posting, you would have to add not:tag:concerns=you there.

The pivoting makes things a little less clear, here are a few more views:

$ hledger bal -N tag:concerns=me   # all three postings inherited `me`
               -30 €  Assets
                20 €  Costs:Food
                10 €  Loaned
$ hledger bal -N tag:concerns=you   # only Loaned has `you`
                10 €  Loaned
$ hledger tags concerns --values  # the `concerns` tag's multiple values
me
you

[nobodyinperson]

tags are multi-valued; they collect and add new values, rather than overriding a single value

Ah, that explains it!

You expect it to override instead, in cases 2a and 2b ?

Yes, I expected child tag specifications to override parent specifications. For all my tags use cases, this is the only sensible behaviour I can think of.

To exclude the posting, you would have to add not:tag:concerns=you there.

This is complicated as there might be very many (unknown) values to exclude.