config: Separate initialization from argument reading.

This lets us fix the bug in __main__ where the exception handler tries to
call my_config.error but my_config wasn't successfully created.

config/hooks: Date hooks configure themselves.

Broadly, this change gets date-specific configuration out of the Config
class, and into the date hooks. Following this pattern, we can have more
specific configuration for more hooks without growing a bunch of custom
methods on the Config class.

There's still more cleanup to be done to finish the job, but this commit is
about as big as I can stand, so this will serve as a good checkpoint.

dynload: Pre-configure objects and skip unconfigured ones.

Now that all importers and hooks receive the configuration object, they can
tell us at initialization time whether or not they actually have enough
configuration to be useful. Initialize them immediately on loading, and
only use importers and hooks that are actually configured.

config: Remove unused method from_section.

The configuration is migrating to have sections set aside for specific
importers and hooks, so using the configuration this way no longer makes
sense.

There's still more cruft to clean from the Config class along the lines of
this change, but that will have to wait until more of the migration is done.

tests: Separate test_hooks_add_entity file.

All the hooks are going to get their own test files over time.

importers: Initialize importers with the user's configuration.

This will enable us to import from sources that require additional
configuration, like web APIs with authentication tokens. This required a
complete revamp of the importer API, but it makes it more uniform with the
hook API, which seems like a plus.

brightfunds: New importer.

This commit adds infrastructure to treat XLS files like CSV files, and
import them using the base classes that already exist for that.

hooks.ledger_entry: Look up templates dynamically.

If there's a 'ledger entry' key in the entry data, use that value as the
name of the template to load. Thanks to this, nbpy2017 could collapse
multiple importers into one.

Otherwise, build a default template name based on the importer source, and
try to use that.

All the configuration names now end with "ledger entry" instead of starting
with "template". This makes it clearer what they're for, in case we
support other kinds of output templates in the future.

I ended up changing the names of some of the importers so the default
template name was nice, rather than specifying template names for all of
them, to reduce the amount of name discrepancies across the codebase.

hooks.ledger_entry: Bring in all Ledger-specific code.

This includes the old "template" module, plus the associated
template-loading code from config.

config: Make get_section a public method.

As hooks and importers both require more configuration, they're gonna need
to have lower-level access to configuration settings.

hooks.ledger_entry: New hook to output Ledger entries.

This is roughly the smallest diff necessary to move output to a hook.
There's a lot of code reorganization that should still happen to bring it
better in line with this new structure.

config: Add open_output_file method.

Make this functionality accessible to hooks.

README: Document Benevity disbursement importer from 1b1e2d0.

With this, we've finished paying down our recent technical debt.

TODO: Format as reST.

Just to reduce the number of things people need to know to contribute.

strparse: Fix default limit argument in split functions.

The underlying string methods don't accept None.

template: Bugfix newline handling.

If a rendering ended with a zeroed out split, the output would contain an
extraneous trailing newline. Rearrange the code to print a newline
preceding each content line.

strparse: Improve extra text parsing in currency_decimal.

This allows a symbol and a currency code to be in different parts of the
string, as long as there's at most one of each.

main: Clean noop line.

I guess this was a remnant when I was trying to implement date filtering
before I decided to do it with hooks.

hooks: run() return value controls processing of entry data.

Instead of using in-band signaling with the entry_data dict.
I don't know why I didn't think of this in the first place.

hooks: Filter entries by setting entry_data['_hook_cancel'] to True.

The current method only works for plain dicts and other simple mappings.
Mapping that may still contain items after .clear(), like ChainMap, can't
rely on the old method.

template: render() takes a single mapping argument.

This lets it receive other mapping types like ChainMap.

util: Rename to strparse.

Better describes what the functions in the module do.

util: Add parse_currency_dec.

The current importers trim lots of extraneous symbols and whitespace from
currency strings before passing them to Decimal(). This function takes care
of all that in a single place.

hooks.invoice_payment: New hook.

In the long run, this should be some kind of configurable hook to run
arbitrary regexps on arbitrary entry fields, and then on a match, add
another field with a format string using the match data. But this is
reasonably generic for now.

main: Open files with errors='replace'.

Stripe payment CSVs have apparently random encoding, and no way what to tell
what they're using beyond chardet.

The better way to fix this is following the existing to-do item, opening
files in binary mode and creating an importer base class to provide text
wrapping. This would let each importer be as strict as possible.

importers.stripe: Generate payment entries for refunded payments.

In order to be refunded, the payment had to be paid in the first place, so
we should still generate an entry for it.

main: Don't use importers whose template isn't configured.

This lets users import only some information from sources that provide
multiple transaction streams at once.

template: Add is_empty method.

This makes it easy to detect when a user has specifically cleared a
template; i.e., from a specific configuration section.

importers.nbpy2017: Yield entries for the original invoice.

This lets us write up a cleaner separation between the original act of
invoicing and the Stripe payment.

importers.nbpy2017: Don't deal with refunds.

Only one refund was actually issued, so we can deal with that manually. But
the code did its job of revealing those cases and checking we were handling
them well.

template: Nicer formatting of account split lines.

This provides nicer output when the whole line can fit in the desired width,
even if the account name or amount are unusually long.

importers: New importer for North Bay Python invoices.

This probably has no use outside Conservancy.

importers: Refactor out a base CSV importer class.

I'm going to build the Stripe importer on top of this.

template: Template amounts are now written with fuller expressions.

This makes it possible to support transactions that divide funds in
ways other than a simple percentage split. For example, there
might be tax withheld, or a flat fee expense on each transaction.

importers: Refactor out a base CSV importer class.

I'm going to build the Stripe importer on top of this.

template: Templates amounts are now written with fuller expressions.

This makes it possible to support transactions that divide funds in
ways other than a simple percentage split. For example, there
might be tax withheld, or a flat fee expense on each transaction.