aboutsummaryrefslogtreecommitdiffstats
path: root/rplugin/python3/deoplete
diff options
context:
space:
mode:
Diffstat (limited to 'rplugin/python3/deoplete')
-rw-r--r--rplugin/python3/deoplete/sources/beancount.py37
1 files changed, 27 insertions, 10 deletions
diff --git a/rplugin/python3/deoplete/sources/beancount.py b/rplugin/python3/deoplete/sources/beancount.py
index c21c057..dcc6a27 100644
--- a/rplugin/python3/deoplete/sources/beancount.py
+++ b/rplugin/python3/deoplete/sources/beancount.py
@@ -1,3 +1,4 @@
+import collections
import re
from deoplete.source.base import Base
@@ -9,8 +10,10 @@ try:
except ImportError:
HAS_BEANCOUNT = False
-DIRECTIVES = ['open', 'close', 'commodity', 'txn', 'balance', 'pad', 'note',
- 'document', 'price', 'event', 'query', 'custom']
+DIRECTIVES = [
+ 'open', 'close', 'commodity', 'txn', 'balance', 'pad', 'note', 'document',
+ 'price', 'event', 'query', 'custom'
+]
class Source(Base):
@@ -22,6 +25,7 @@ class Source(Base):
self.mark = '[bc]'
self.filetypes = ['beancount']
self.min_pattern_length = 0
+ self.attributes = collections.defaultdict(list)
def on_init(self, context):
if not HAS_BEANCOUNT:
@@ -35,18 +39,29 @@ class Source(Base):
return m.start() if m else -1
def gather_candidates(self, context):
+ attrs = self.attributes
if re.match(r'^\d{4}[/-]\d\d[/-]\d\d \w*$', context['input']):
return [{'word': x, 'kind': 'directive'} for x in DIRECTIVES]
+ # line that starts with whitespace (-> accounts)
+ if re.match(r'^(\s)+\w+$', context['input']):
+ return [{'word': x, 'kind': 'account'} for x in attrs['accounts']]
+ # directive followed by account
+ if re.match(r'(balance|document|note|open|close|pad(\s\S*)?)\s\w+$',
+ context['input']):
+ return [{'word': x, 'kind': 'account'} for x in attrs['accounts']]
if not context['complete_str']:
return []
first = context['complete_str'][0]
if first == '#':
- return [{'word': '#' + w, 'kind': 'tag'} for w in self._tags]
+ return [{'word': '#' + w, 'kind': 'tag'} for w in attrs['tags']]
elif first == '^':
- return [{'word': '^' + w, 'kind': 'link'} for w in self._links]
+ return [{'word': '^' + w, 'kind': 'link'} for w in attrs['links']]
elif first == '"':
- return [{'word': '"{}"'.format(w), 'kind': 'payee'} for w in self._payees]
- return [{'word': x, 'kind': 'account'} for x in self._accounts]
+ return [{
+ 'word': '"{}"'.format(w),
+ 'kind': 'payee'
+ } for w in attrs['payees']]
+ return []
def __make_cache(self, context):
accounts = set()
@@ -69,7 +84,9 @@ class Source(Base):
if hasattr(entry, 'tags') and entry.tags:
tags.update(entry.tags)
- self._accounts = sorted(accounts)
- self._links = sorted(links)
- self._payees = sorted(payees)
- self._tags = sorted(tags)
+ self.attributes = {
+ 'accounts': sorted(accounts),
+ 'links': sorted(links),
+ 'payees': sorted(payees),
+ 'tags': sorted(tags),
+ }