Comment author: Bradley M. Kuhn (@bkuhn)

This fails:
2013-11-18 Test
Expenses -625.00 EUR {=$1.3262}
Expenses -896.15 EUR {=$1.3348}
Accrued $828.88
Accrued $1,196.18

with:
Unbalanced remainder is:
$2,025.06
-625.00 EUR {=$1.3262}
-896.15 EUR {=$1.3348}
Amount to balance against:
$2,025.06

This works:

2013-11-18 Test
Expenses -625.00 EUR {=$1.3262} @ $1.3262
Expenses -896.15 EUR {=$1.3348} @ $1.3348
Accrued $828.88
Accrued $1,196.18

as does this:

2013-11-18 test2
Expenses $-34.60
Expenses -775.51 CAD {=$0.9524} @ $0.9524
Accrued $773.20

But merged together:

2013-11-18 Test
Expenses -625.00 EUR {=$1.3262} @ $1.3262
Expenses -896.15 EUR {=$1.3348} @ $1.3348
Accrued $828.88
Accrued $1,196.18
Expenses $-34.60
Expenses -775.51 CAD {=$0.9524} @ $0.9524
Accrued $773.20

I get:
Unbalanced remainder is:
$0.01
Amount to balance against:
$2,798.26

What gives?

Comment author: Tripun (@tripun)

Created attachment 96
test input data

Round off occur when the amount of commodity in a transaction has less precision than the calculation involved.
Evaluation is as below:
"
This fails:
2013-11-18 Test
Expenses -625.00 EUR {=$1.3262}
Expenses -896.15 EUR {=$1.3348}
Accrued $828.88
Accrued $1,196.18

with:
Unbalanced remainder is:
$2,025.06
-625.00 EUR {=$1.3262}
-896.15 EUR {=$1.3348}
Amount to balance against:
$2,025.06
"
Expenses are non-zero but cannot be converted to $ which is the unit of display and hence unbalanced amount is what is available in dollars.

==============
"
2013-11-18 Test
Expenses -625.00 EUR {=$1.3262} @ $1.3262
Expenses -896.15 EUR {=$1.3348} @ $1.3348
Accrued $828.88
Accrued $1,196.18
"
After conversion balance is $0.003(approx), the reporting precision is 2 digits since the amounts(quantity) involved have maximum of 2 digit precision. After round off the balance is zero and hence there is no unbalanced amount and the test passes.

=====================
"

2013-11-18 test2
Expenses $-34.60
Expenses -775.51 CAD {=$0.9524} @ $0.9524
Accrued $773.20

"
Same explanation as above. round off leads to zero unbalanced amount.

=====================================
"
2013-11-18 Test
Expenses -625.00 EUR {=$1.3262} @ $1.3262
Expenses -896.15 EUR {=$1.3348} @ $1.3348
Accrued $828.88
Accrued $1,196.18
Expenses $-34.60
Expenses -775.51 CAD {=$0.9524} @ $0.9524
Accrued $773.20
"

Attached file: roundoff.dat (text/plain, 791 bytes)
Description: test input data

Comment author: Tripun (@tripun)

"
2013-11-18 Test
Expenses -625.00 EUR {=$1.3262} @ $1.3262
Expenses -896.15 EUR {=$1.3348} @ $1.3348
Accrued $828.88
Accrued $1,196.18
Expenses $-34.60
Expenses -775.51 CAD {=$0.9524} @ $0.9524
Accrued $773.20
"

After conversion balance is $0.008(approx), the reporting precision is 2 digits since the amounts(quantity) involved have maximum of 2 digit precision. After round off the balance is $0.01 and hence there is unbalanced amount and the test fails.

Comment author: John Wiegley (@jwiegley)

Bradley, you have a point that if the @ version works, the {=} version should
work equally well. I'll make a separate bug for that.

Comment author: Martin Michlmayr (@tbm)

(In reply to comment #3)

Bradley, you have a point that if the @ version works, the {=} version
should work equally well. I'll make a separate bug for that.

I believe it's already covered by bug BZ#630.

Comment author: Martin Michlmayr (@tbm)

I'm changing this to an enhancement request: ledger needs support for rounding. As Tripun pointed out, there is no bug as such. It's just that ledger keeps a lot of precision internally and this can lead to errors.

In most cases, we want to round to 2 digits. There should be a way for ledger to do such rounding.

Comment author: Martin Michlmayr (@tbm)

(In reply to comment #2)

"
2013-11-18 Test
Expenses -625.00 EUR {=$1.3262} @ $1.3262
Expenses -896.15 EUR {=$1.3348} @ $1.3348
Accrued $828.88
Accrued $1,196.18
Expenses $-34.60
Expenses -775.51 CAD {=$0.9524} @ $0.9524
Accrued $773.20
"

After conversion balance is $0.008(approx), the reporting precision is 2
digits since the amounts(quantity) involved have maximum of 2 digit
precision. After round off the balance is $0.01 and hence there is
unbalanced amount and the test fails.

Yes, and you can see this if you put the following in the file:

D $1000.000

Unbalanced remainder is:
$0.008
Amount to balance against:
$2,798.260

Comment author: Martin Michlmayr (@tbm)

So here's an interesting example which shows that you can get different results to what you'd expect because of ledger keeping so much precision.

Using a lot of precision using figures like @ $1.3762 gives us:

2014-01-01 * Expense in EUR
Accounts Payable -€154.90 {=$1.3762} @ $1.3762
Expenses:foo

2014-02-01 * Person paid 4338.00 CZK, not 154.90 EUR
Accounts Payable €154.90 {=$1.3762} @ $1.3762
Accounts Payable -4338.00 CZK {=$0.05026} @ $0.05026
Expenses:Currency $4.85

Note that Expenses:Currency is $4.85.

But now let's use amounts rounded to 2 digits:

2014-01-01 * Expense in EUR
Accounts Payable -€154.90 {{=$213.17}} @ $1.3762
Expenses:foo

2014-02-01 * Person paid 4338.00 CZK, not 154.90 EUR
Accounts Payable €154.90 {{=$213.17}} @ @ $213.17
Accounts Payable -4338.00 CZK {{=$218.03}} @ @ $218.03 ; $0.05026
Expenses:Currency $4.86

Now Expenses:Currency is $4.86.

Why the difference?

With full precision:

154.90 * 1.3762 = 213.17338
-4338.00 * 0.05026 = -218.02788
213.17338 - 218.02788 = -4.8545

4.8545, which is rounded to 4.85 at display precision 2.

On the other hand, using the amounts that were rounded to 2 digits fo precision:

213.17 - 218.03 = -4.86

Obviously, both solutions are correct.

But there should be a way to specify a rounding precision, so you can have the first example (i.e. exchange amounts specified as @ $1.3762) and still get an Expenses:Currency of $4.86.

Comment author: Martin Michlmayr (@tbm)

Another request for rounding is discussed in #781 (with a good test case).