This broke in 1.43.
Now we ignore any IOException whose message contains "broken pipe".
Hopefully this matches pre-1.43 behaviour and doesn't hide real errors.
This upgrades Account to enable it to store a multiperiod balance, with
a separate balance for each date period. This enables it do the hard
work in MultiBalanceReport.
Some new types are created to enable convenient operation of accounts.
- `BalanceData` is a type which stores an exclusive balance, inclusive
balance, and number of postings. This was previously directly stored
in Account, but is now factored into a separate data type.
- `PeriodData` is a container which stores date-indexed data, as well as
pre-period data. In post cases, this represents the report spans,
along with the historical data.
- Account becomes polymorphic, allowing customisation of the type of
data it stores. This will usually be `BalanceData`, but in
`BudgetReport` it can use `These BalanceData BalanceData` to store
both actuals and budgets in the same structure. The data structure
changes to contain a `PeriodData`, allowing multiperiod accounts.
Some minor changes are made to behaviour for consistency:
- --declared treats parent accounts consistently.
- --flat --empty ensures that implied accounts with no postings are not displayed, but
accounts with zero balance and actual postings are.
This and/or the previous related change also changed a few cases which
always traced or always logged; now all debug output is consistently
either traced or logged.
A very long-awaited improvement: for unicode exceptions, and I/O
exceptions which look like they were caused by a unicode error
(usually text decoding failure), our error message now includes
an explanation and advice on what to do.
Currently this uses the GHC.IO.Encoding API, which is not ideal:
"The API of this module is unstable and not meant to be consumed by
the general public. If you absolutely must depend on it, make sure to
use a tight upper bound, e.g., base < 4.X rather than base < 5,
because the interface can change rapidly without much warning."
Also it relies on scanning for patterns in GHC's various
unicode-related error messages, which may not be complete and could
change in future. To do: try the encoding package's IO helpers,
perhaps they give more specific exceptions.
