let s:using_python3 = has('python3')
" Equivalent to python's startswith
" Matches based on user's ignorecase preference
function! s:startswith(string, prefix) abort
return strpart(a:string, 0, strlen(a:prefix)) == a:prefix
endfunction
" Align currency on decimal point.
function! beancount#align_commodity(line1, line2) abort
" Save cursor position to adjust it if necessary.
let l:cursor_col = col('.')
let l:cursor_line = line('.')
" Increment at start of loop, because of continue statements.
let l:current_line = a:line1 - 1
while l:current_line < a:line2
let l:current_line += 1
let l:line = getline(l:current_line)
" This matches an account name followed by a space in one of the two
" following cases:
" - A posting line, i.e., the line starts with indentation followed
" by an optional flag and the account.
" - A balance directive, i.e., the line starts with a date followed
" by the 'balance' keyword and the account.
" - A price directive, i.e., the line starts with a date followed by
" the 'price' keyword and a currency.
let l:end_account = matchend(l:line, '^\v' .
\ '[\-/[:digit:]]+\s+balance\s+([A-Z][A-Za-z0-9\-]+)(:[A-Z][A-Za-z0-9\-]*)+ ' .
\ '|[\-/[:digit:]]+\s+price\s+\S+ ' .
\ '|\s+([!&#?%PSTCURM]\s+)?([A-Z][A-Za-z0-9\-]+)(:[A-Z][A-Za-z0-9\-]*)+ '
\ )
if l:end_account < 0
continue
endif
" Where does the number begin?
let l:begin_number = matchend(l:line, '^ *', l:end_account)
" Look for a minus sign and a number (possibly containing commas) and
" align on the next column.
let l:separator = matchend(l:line, '^\v([-+])?[,[:digit:]]+', l:begin_number) + 1
if l:separator < 0 | continue | endif
let l:has_spaces = l:begin_number - l:end_account
let l:need_spaces = g:beancount_separator_col - l:separator + l:has_spaces
if l:need_spaces < 0 | continue | endif
call setline(l:current_line, l:line[0 : l:end_account - 1] . repeat(' ', l:need_spaces) . l:line[ l:begin_number : -1])
if l:current_line == l:cursor_line && l:cursor_col >= l:end_account
" Adjust cursor position for continuity.
call cursor(0, l:cursor_col + l:need_spaces - l:has_spaces)
endif
endwhile
endfunction
function! s:count_expression(text, expression) abort
return len(split(a:text, a:expression, 1)) - 1
endfunction
function! s:sort_accounts_by_depth(name1, name2) abort
let l:depth1 = s:count_expression(a:name1, ':')
let l:depth2 = s:count_expression(a:name2, ':')
return l:depth1 == l:depth2 ? 0 : l:depth1 > l:depth2 ? 1 : -1
endfunction
let s:directives = ['open', 'close', 'commodity', 'txn', 'balance', 'pad', 'note', 'document', 'price', 'event', 'query', 'custom']
" ------------------------------
" Completion functions
" ------------------------------
function! beancount#complete(findstart, base) abort
if a:findstart
let l:col = searchpos('\s', 'bn', line('.'))[1]
if l:col == 0
return -1
else
return l:col
endif
endif
let l:partial_line = strpart(getline('.'), 0, getpos('.')[2]-1)
" Match directive types
if l:partial_line =~# '^\d\d\d\d\(-\|/\)\d\d\1\d\d $'
return beancount#complete_basic(s:directives, a:base, '')
endif
" If we are using python3, now is a good time to load everything
call beancount#load_everything()
" Split out the first character (for cases where we don't want to match the
" leading character: ", #, etc)
let l:first = strpart(a:base, 0, 1)
let l:rest = strpart(a:base, 1)
if l:partial_line =~# '^\d\d\d\d\(-\|/\)\d\d\1\d\d event $' && l:first ==# '"'
return beancount#complete_basic(b:beancount_events, l:rest, '"')
endif
let l:two_tokens = searchpos('\S\+\s', 'bn', line('.'))[1]
let l:prev_token = strpart(getline('.'), l:two_tokens, getpos('.')[2] - l:two_tokens)
" Match curriences if previous token is number
if l:prev_token =~# '^\d\+\([\.,]\d\+\)*'
call beancount#load_currencies()
return beancount#complete_basic(b:beancount_currencies, a:base, '')
endif
if l:first ==# '#'
call beancount#load_tags()
return beancount#complete_basic(b:beancount_tags, l:rest, '#')
elseif l:first ==# '^'
call beancount#load_links()
return beancount#complete_basic(b:beancount_links, l:rest, '^')
elseif l:first ==# '"'
call beancount#load_payees()
return beancount#complete_basic(b:beancount_payees, l:rest, '"')
else
call beancount#load_accounts()
return beancount#complete_account(a:base)
endif
endfunction
function! beancount#get_root() abort
if exists('b:beancount_root')
return b:beancount_root
endif
return expand('%')
endfunction
function! beancount#load_everything() abort
if s:using_python3 && !exists('b:beancount_loaded')
let l:root = beancount#get_root()
python3 << EOF
import vim
from beancount import loader
from beancount.core import data
accounts = set()
currencies = set()
events = set()
links = set()
payees = set()
tags = set()
entries, errors, options_map = loader.load_file(vim.eval('l:root'))
for index, entry in enumerate(entries):
if isinstance(entry, data.Open):
accounts.add(entry.account)
if entry.currencies:
currencies.update(entry.currencies)
elif isinstance(entry, data.Commodity):
currencies.add(entry.currency)
elif isinstance(entry, data.Event):
events.add(entry.type)
elif isinstance(entry, data.Transaction):
if entry.tags:
tags.update(entry.tags)
if entry.links:
links.update(entry.links)
if entry.payee:
payees.add(entry.payee)
vim.command('let b:beancount_accounts = [{}]'.format(','.join(repr(x) for x in sorted(accounts))))
vim.command('let b:beancount_currencies = [{}]'.format(','.join(repr(x) for x in sorted(currencies))))
vim.command('let b:beancount_events = [{}]'.format(','.join(repr(x) for x in sorted(events))))
vim.command('let b:beancount_links = [{}]'.format(','.join(repr(x) for x in sorted(links))))
vim.command('let b:beancount_payees = [{}]'.format(','.join(repr(x) for x in sorted(payees))))
vim.command('let b:beancount_tags = [{}]'.format(','.join(repr(x) for x in sorted(tags))))
vim.command('let b:beancount_loaded = v:true'.format(','.join(repr(x) for x in sorted(tags))))
EOF
endif
endfunction
function! beancount#load_accounts() abort
if !s:using_python3 && !exists('b:beancount_accounts')
let l:root = beancount#get_root()
let b:beancount_accounts = beancount#query_single(l:root, 'select distinct account;')
endif
endfunction
function! beancount#load_tags() abort
if !s:using_python3 && !exists('b:beancount_tags')
let l:root = beancount#get_root()
let b:beancount_tags = beancount#query_single(l:root, 'select distinct tags;')
endif
endfunction
function! beancount#load_links() abort
if !s:using_python3 && !exists('b:beancount_links')
let l:root = beancount#get_root()
let b:beancount_links = beancount#query_single(l:root, 'select distinct links;')
endif
endfunction
function! beancount#load_currencies() abort
if !s:using_python3 && !exists('b:beancount_currencies')
let l:root = beancount#get_root()
let b:beancount_currencies = beancount#query_single(l:root, 'select distinct currency;')
endif
endfunction
function! beancount#load_payees() abort
if !s:using_python3 && !exists('b:beancount_payees')
let l:root = beancount#get_root()
let b:beancount_payees = beancount#query_single(l:root, 'select distinct payee;')
endif
endfunction
" General completion function
function! beancount#complete_basic(input, base, prefix) abort
let l:matches = filter(copy(a:input), 's:startswith(v:val, a:base)')
return map(l:matches, 'a:prefix . v:val')
endfunction
" Complete account name.
function! beancount#complete_account(base) abort
if g:beancount_account_completion ==? 'chunks'
let l:pattern = '^\V' . substitute(a:base, ':', '\\[^:]\\*:', 'g') . '\[^:]\*'
else
let l:pattern = '^\V\.\*' . substitute(a:base, ':', '\\.\\*:\\.\\*', 'g') . '\.\*'
endif
let l:matches = []
let l:index = -1
while 1
let l:index = match(b:beancount_accounts, l:pattern, l:index + 1)
if l:index == -1 | break | endif
call add(l:matches, matchstr(b:beancount_accounts[l:index], l:pattern))
endwhile
if g:beancount_detailed_first
let l:matches = reverse(sort(l:matches, 's:sort_accounts_by_depth'))
endif
return l:matches
endfunction
function! beancount#query_single(root_file, query) abort
python << EOF
import vim
import subprocess
import os
# We intentionally want to ignore stderr so it doesn't mess up our query processing
output = subprocess.check_output(['bean-query', vim.eval('a:root_file'), vim.eval('a:query')], stderr=open(os.devnull, 'w')).split('\n')
output = output[2:]
result_list = [y for y in (x.strip() for x in output) if y]
vim.command('return [{}]'.format(','.join(repr(x) for x in sorted(result_list))))
EOF
endfunction
" Call bean-doctor on the current line and dump output into a scratch buffer
function! beancount#get_context() abort
let l:context = system('bean-doctor context ' . expand('%') . ' ' . line('.'))
botright new
setlocal buftype=nofile bufhidden=hide noswapfile
call append(0, split(l:context, '\v\n'))
normal! gg
endfunction