1.29: Valid entries with equity conversion postings are rejected if hledger can't detect balancing postings

[the-solipsist]

This entry is parsed without issue with hledger bal in v1.28:

2017-12-08 Giordano
    Expenses:Gift:XX-XXXXXXX                 HKD 118.00
    Expenses:Personal:Clothing               HKD 118.00
    Equity:Trading:Currency:INR-HKD:HKD     HKD -236.00
    Equity:Trading:Currency:INR-HKD:INR       ₹2,150.77
    Liabilities:Credit-Card:XXX              ₹-2,150.77

However, this is no longer the case with v1.29+, which throws up the error:

There is not a unique posting which matches the conversion posting pair:
    Equity:Trading:Currency:INR-HKD:INR       ₹2,150.77
    Equity:Trading:Currency:INR-HKD:HKD     HKD -236.00

This is essentially doing a strict --infer-costs check, rather than / in addition to whatever checks the journal parsing code was doing earlier.

This works fine:

2017-12-08 Giordano
    Expenses:Personal:Clothing               HKD 236.00
    Equity:Trading:Currency:INR-HKD:HKD     HKD -236.00
    Equity:Trading:Currency:INR-HKD:INR       ₹2,150.77
    Liabilities:Credit-Card:XXX              ₹-2,150.77
    Expenses:Gift:XX-XXXXXXX                 HKD 118.00
    Expenses:Personal:Clothing              HKD -118.00

[the-solipsist]

Strangely, I don't recall having had an issue with --infer-costs earlier, though the 5-6 problematic entries I still had remaining should have meant that --infer-costs wouldn't work. At any rate, I fixed those entries, and things are working again. So I'm actually thankful for this 'bug'. But given that journal parsing isn't supposed to be this strict, I think it should still qualify as a bug.

[simonmichael]

The problem occurs during this new pass that is done to allow redundant costs + conversion postings:

hledger/hledger-lib/Hledger/Read/Common.hs

Line 334 in c2c668e

>>= journalMarkRedundantCosts -- Mark redundant costs, to help journalBalanceTransactions ignore them

in addCostsToPostings, when dryrun' is true:

hledger/hledger-lib/Hledger/Data/Transaction.hs

Line 298 in c2c668e

| otherwise -> Left "There is not a unique posting which matches the conversion posting pair:"

If I make that case do nothing instead, it fixes @the-solipsist's example but breaks the three tests starting at costs test 39:

hledger/hledger/test/journal/costs.test

Line 510 in c2c668e

# 39. If a conversion pair matches several postings it should throw an error

I for one need a fresh mind to see when exactly it should raise an error here.

[simonmichael]

I have pushed a fix (3357c27, meant to push a PR instead..). This changes the transaction-balancing and --infer-cost behaviour of 1.29 and 1.30, to be more forgiving, doing nothing rather than raise errors, which solves the present issue and I think will not cause problems, and is consistent with --infer-equity, infer-market-prices, -B and -V.