`check` command should not stop after the first error

[pwseo]

Currently, if one turns on strict mode or uses the check command (eg. check accounts, check payees), hledger stops after the first undeclared account/payee is encountered.

While this is technically correct, it makes one repeat the edit-check-fix cycle more times than should be needed.

Here's an example:

$ hledger --version
hledger 1.34, linux-x86_64
$ cat a.j
2024-06-08 A | Txn 1
    Undeclared:One      +5 EUR
    Undeclared:Another  -5 EUR

2024-06-08 B | Txn 2
    Undeclared:Another  +5 EUR
    Undeclared:One      -5 EUR
$ hledger -f a.j accounts
Undeclared:Another
Undeclared:One
$ hledger -f a.j payees
A
B

As it stands, we have 2 undeclared accounts and 2 undeclared payees. Let's check them with hledger:

$ hledger -f a.j check accounts
hledger: Error: /tmp/tmp.EQDr9awkqd/a.j:2:
  | 2024-06-08 A | Txn 1
2 |     Undeclared:One               5 EUR
  |     ^^^^^^^^^^^^^^
  |     Undeclared:Another          -5 EUR

Strict account checking is enabled, and
account "Undeclared:One" has not been declared.
Consider adding an account directive. Examples:

account Undeclared:One
account Undeclared:One    ; type:A  ; (L,E,R,X,C,V)

The same thing happens for payees: hledger never sees payee B and thus never complains about it because it quits after reporting A is not declared.

I think it would be helpful if hledger continued processing the data and collected as much undeclared accounts / payees as possible before presenting them to the user. Perhaps this would not belong in the -s argument's default behaviour (so as not to overwhelm new users), but only in the explicit check command (it does make sense to report all undeclared accounts if I ask to check accounts, I think).

[simonmichael]

I'd be happy to test a PR exploring this. Some checks are relatively easy to continue from on failure; others are not.

But I don't find it causing pain in practice, do you ? I don't have a slow edit-check-fix cycle.

[pwseo]

I'd be happy to test a PR exploring this. Some checks are relatively easy to continue from on failure; others are not.

Maybe one day. Right now my Haskell skills are subpar and I have no familiarity with hledger's code base (sadly). I'll try and change that in the future.

But I don't find it causing pain in practice, do you ? I don't have a slow edit-check-fix cycle.

In a carefully maintained chart of accounts this isn't problematic (especially if one has some kind of auto-complete for account names), but scenarios vary. I recently started tracking my finances with hledger and there was a lot of checking and fixing to do in the beginning (because my chart of accounts wasn't fully developed).

Still, on a more general note, I think having the check command try and report as many errors as it can is a positive change, while passing the --strict flag should still quit after the first error is found.

[simonmichael]

Not detracting from your point, but remember strict account checking is optional and simply avoiding it is another option when the chart of accounts hasn't yet stabilised.

[pwseo]

Yes, I know, but strictness checking is also the only way to know which accounts have not been declared and are being used. One can always get the used accounts' listing from a non-strict hledger accounts, but it has its downsides.

Thank you for the suggestion, though.