diff options
author | tpearson <tpearson@283d02a7-25f6-0310-bc7c-ecb5cbfe19da> | 2011-07-04 22:38:03 +0000 |
---|---|---|
committer | tpearson <tpearson@283d02a7-25f6-0310-bc7c-ecb5cbfe19da> | 2011-07-04 22:38:03 +0000 |
commit | dadc34655c3ab961b0b0b94a10eaaba710f0b5e8 (patch) | |
tree | 99e72842fe687baea16376a147619b6048d7e441 /contrib/csvbankingqif.py | |
download | kmymoney-dadc34655c3ab961b0b0b94a10eaaba710f0b5e8.tar.gz kmymoney-dadc34655c3ab961b0b0b94a10eaaba710f0b5e8.zip |
Added kmymoney
git-svn-id: svn://anonsvn.kde.org/home/kde/branches/trinity/applications/kmymoney@1239792 283d02a7-25f6-0310-bc7c-ecb5cbfe19da
Diffstat (limited to 'contrib/csvbankingqif.py')
-rwxr-xr-x | contrib/csvbankingqif.py | 151 |
1 files changed, 151 insertions, 0 deletions
diff --git a/contrib/csvbankingqif.py b/contrib/csvbankingqif.py new file mode 100755 index 0000000..4e0d7ea --- /dev/null +++ b/contrib/csvbankingqif.py @@ -0,0 +1,151 @@ +#!/usr/bin/env python +# -*- coding: iso-8859-1 -*- +# +#*************************************************************************** +# csvbankingqif.py - description +# ------------------- +# begin : Sat 31 Oct. 2009 +# copyright : (C) 2009 by Allan Anderson +# email : aganderson@ukonline.co.uk +# +#***************************************************************************/ +# +#*************************************************************************** +#* * +#* This program is free software; you can redistribute it and/or modify * +#* it under the terms of the GNU General Public License as published by * +#* the Free Software Foundation; either version 2 of the License, or * +#* (at your option) any later version. * +#* * +#***************************************************************************/ +# +# *** NOTE *** +# It may be necessary to remove the second line, before running. +# It may be necessary also, to change the currency symbol if your file +# includes one. +# +# Simple utility to convert a csv format file, as from a bank, to qif +# format for KMyMoney2. There is no standard for the layout of such a +# file, but generally there will be a header line which indicates the +# layout of the fields within the file. Even then though, the order of +# the columns may vary. It is assumed, though, that the first column +# will contain the date, in 'dd MM yy' format, 'MM' being the month +# name or number. +# The second column is the detail. The third and fourth columns are +# assumed to be debits and credits. Even fron the same bank, these +# columns may be reversed, but the script handles this. Alternatively, +# the third column may be the amount. There may also be additional +# columns, such as current balance, but these are all ignored. +# Apart from the header line, there are likely to be other lines, with +# account number, balance details, etc. These are skipped. +# +# First, make the script executable: chmod u+x csvbankinyqif.py . +# The script should be added to the KMM QIF import profile. In KMM, open +# Tools/QIF Profile Editor and click on 'New' at the bottom. then enter a +# name, such as csvbank, then click 'OK'. Next, click on that name in the +# next window, and open the Filter tab. For the 'Input filter location', +# select the location you chose for the script file. For the Input filter +# file type, enter *.csv, or whatever extension your data file has. +# Finally, click 'OK'. +# When ready, select File/Import/QIF, and browse to your data file, then +# select your new filter profile, and click 'Import'. +# + +#desc="date","detail","debit","credit" +mnths=['Jan','Feb','Mar','Apr','May','Jun','Jul','Aug','Sep','Oct','Nov','Dec'] +currency = '£' +setup = True +debsfirst = False # credit column is before debit +both = False # only one amount column +print("!Type:Bank") +while 1: + try: + line=raw_input() + except:break + if line == "" : continue # empty line + line = line.replace('"','',) # Strip out ' "' quotes + line = line.replace(currency,'',) # Strip out '£', etc. symbol + cols = line.split(',') # Seperator between columns + if setup: +# +# *** SETUP *** +# + dt = cols[0][0:2] # Might be a date (day) + datefound = ((dt > '0') and (dt < '32')) #this looks like a date + hdrfound = (cols[0] == 'Date') + if not datefound and not hdrfound: continue# still in hdrs + if hdrfound: +# +# *** 'Date', so now in header *** +# + hdrfound = False + #line = line.replace(' ','',) # Strip out spaces in hdr + cols[2] = cols[2].replace(' ','',) # Strip out spaces in hdr + if cols[2] == 'Debits': + debsfirst = True + continue + elif cols[2] == 'Credits': + debsfirst = False + continue + elif cols[2] == 'Amount': + both = True + continue + else: + print 'Error in col[2]' + print '*** Error in header - col 2 s/b Debit, Credit, or Amount' + #continue + exit + setup ==False +# +# *** Transactions *** +# + cnum = 0 # First column + for col in cols: + if cnum > 3: break +# +# # Process Date +# + elif cnum == 0: + col =col.replace(' ','/',2) # Change date seperator to '/' + m = col.split('/') +# *** Check if month not numeric + mn = m[1][0:3] # Extract month string from field 2 + fld = 2 + try: + mnth = mnths.index(mn) # Get month number + except ValueError: # Field 2 not a valid month name + mn = m[0][0:3] # .. so try field 1 + fld = 1 + try: + mnth = mnths.index(mn) + except ValueError: # Nor is field 1 + dat = ''.join(col) # ..so use as is (numeric) + else: # Field 1 is month name + dat = col[1:3] + str(mnth + 1) + '/' +m[2] + else: # Field 2 is month name + dat = col[0:3] + str(mnth + 1) + '/' +m[2] + line = 'D' + dat+'\n' +# +# # Detail column +# + elif cnum == 1: + #col = col.replace('"','') + line = line + 'P' + col +'\n' +# +# # Debit or credit column +# + elif cnum == 2: + if col != "": + if debsfirst == True: # This is Debit column + col = '-' + col # Mark as -ve + line = line + 'T' + col +'\n' +# +# # Credit or debit? +# + elif ((cnum == 3) and (both == False)): + if col != "": + if ((debsfirst == False) ): + col = '-' + col # Mark as -ve + line = line + 'T' + col + '\n' + cnum+=1 + print line + '^' # output this entry |