191 lines
6.7 KiB
Python
191 lines
6.7 KiB
Python
#
|
|
# Python script to generate a BOM from a KiCad generic netlist
|
|
#
|
|
# Standard for Adapteva Schematics
|
|
#
|
|
# Required attributes: Manufacturer, MFRPN, Description (in DCM file)
|
|
|
|
from __future__ import print_function
|
|
|
|
# Import the KiCad python helper module and the csv formatter
|
|
import kicad_netlist_reader
|
|
import csv
|
|
import sys
|
|
|
|
if len(sys.argv) != 3 and len(sys.argv) != 4:
|
|
print("Usage ", __file__, "<generic_netlist.xml> <output.csv> [all|coll|each]", file=sys.stderr)
|
|
sys.exit(1)
|
|
|
|
type = "coll"
|
|
if len(sys.argv) == 4:
|
|
type = sys.argv[3]
|
|
|
|
def myEqu(self, other):
|
|
"""myEqu is a more advanced equivalence function for components which is
|
|
used by component grouping. Normal operation is to group components based
|
|
on their Value, Library source, and Library part.
|
|
|
|
Here we also compare the custom fields Manufacturer, Mfr P/N, and "DNI"
|
|
as well as the assigned footprint. If these fields are not used in some
|
|
parts they will simply be ignored (they will match as both will be empty
|
|
strings).
|
|
|
|
"""
|
|
result = True
|
|
if self.getValue() != other.getValue():
|
|
result = False
|
|
elif self.getLibName() != other.getLibName():
|
|
result = False
|
|
elif self.getPartName() != other.getPartName():
|
|
result = False
|
|
elif self.getFootprint() != other.getFootprint():
|
|
result = False
|
|
elif self.getField("Manufacturer") != other.getField("Manufacturer"):
|
|
result = False
|
|
elif self.getField("MFRPN") != other.getField("MFRPN"):
|
|
result = False
|
|
elif self.getField("DNI") != other.getField("DNI"):
|
|
result = False
|
|
return result
|
|
|
|
# Override the component equivalence operator - it is important to do this
|
|
# before loading the netlist, otherwise all components will have the original
|
|
# equivalency operator. NOTE this is "eq" not "equ"!
|
|
kicad_netlist_reader.comp.__eq__ = myEqu
|
|
|
|
# Generate an instance of a generic netlist, and load the netlist tree from
|
|
# the command line option. If the file doesn't exist, execution will stop
|
|
net = kicad_netlist_reader.netlist(sys.argv[1])
|
|
|
|
# Open a file to write to, if the file cannot be opened report error
|
|
try:
|
|
f = open(sys.argv[2], 'w')
|
|
except IOError:
|
|
print(__file__, ":", e, file=sys.stderr)
|
|
sys.exit(2)
|
|
|
|
# subset the components to those wanted in the BOM, controlled
|
|
# by blacklists
|
|
|
|
kicad_netlist_reader.excluded_references = [
|
|
'TP[0-9]+',
|
|
'MH[0-9]+',
|
|
'FID[0-9]+'
|
|
]
|
|
|
|
components = net.getInterestingComponents()
|
|
|
|
compfields = net.gatherComponentFieldUnion(components)
|
|
partfields = net.gatherLibPartFieldUnion()
|
|
|
|
columnset = compfields | partfields # union
|
|
|
|
# remove Reference, Value, Datasheet, and Footprint, they will come from 'columns' below
|
|
columnset -= set( ['Reference', 'Value', 'Footprint', 'Manufacturer', 'MFRPN', 'DNI'] )
|
|
|
|
# prepend an initial 'hard coded' list and put the enchilada into list 'columns'
|
|
columns = ['Item', 'Qty', 'Reference(s)', 'DNI', 'MFRPN', 'Description', 'Manufacturer',
|
|
'Lib', 'Part Name', 'Footprint', 'Datasheet'] + sorted(list(columnset))
|
|
|
|
# Create a new csv writer object to use as the output formatter
|
|
out = csv.writer(f, lineterminator='\n', delimiter=',', quotechar='\"', quoting=csv.QUOTE_MINIMAL)
|
|
|
|
# override csv.writer's writerow() to support utf8 encoding:
|
|
def writerow( acsvwriter, columns ):
|
|
utf8row = []
|
|
for col in columns:
|
|
utf8row.append( str(col).encode('utf8') )
|
|
acsvwriter.writerow( utf8row )
|
|
|
|
# Output a set of rows as a header providing general information
|
|
writerow( out, ['Source:', net.getSource()] )
|
|
writerow( out, ['Date:', net.getDate()] )
|
|
writerow( out, ['Tool:', net.getTool()] )
|
|
writerow( out, ['Component Count:', len(components)] )
|
|
writerow( out, [] )
|
|
|
|
if type == "all" or type == "each":
|
|
|
|
print( "Generating per-item list..." )
|
|
writerow( out, ['Individual Components:'] )
|
|
writerow( out, [] ) # blank line
|
|
writerow( out, columns )
|
|
|
|
# Output all the interesting components individually first:
|
|
row = []
|
|
for c in components:
|
|
del row[:]
|
|
row.append('') # item is blank in individual table
|
|
row.append('') # Qty is always 1, why print it
|
|
row.append( c.getRef() ) # Reference
|
|
row.append( c.getField("DNI") ) # DNI
|
|
row.append( c.getField("MFRPN") ) # Mfr's Part #
|
|
row.append( c.getDescription() )
|
|
row.append( c.getField("Manufacturer") )
|
|
row.append( c.getLibName() )
|
|
row.append( c.getPartName() )
|
|
row.append( c.getFootprint() )
|
|
row.append( c.getDocs() )
|
|
|
|
# from column 11 upwards, use the fieldnames to grab the data
|
|
for field in columns[11:]:
|
|
row.append( c.getField( field ) );
|
|
|
|
writerow( out, row )
|
|
|
|
writerow( out, [] ) # blank line
|
|
writerow( out, [] ) # blank line
|
|
writerow( out, [] ) # blank line
|
|
|
|
if type == "all" or type == "coll":
|
|
|
|
print( 'Generating collated list...' )
|
|
writerow( out, ['Collated Components:'] )
|
|
writerow( out, [] ) # blank line
|
|
writerow( out, columns ) # reuse same columns
|
|
|
|
# Get all of the components in groups of matching parts + values
|
|
# (see kicad_netlist_reader.py)
|
|
grouped = net.groupComponents(components)
|
|
|
|
# Output component information organized by group, aka as collated:
|
|
item = 0
|
|
row = []
|
|
for group in grouped:
|
|
del row[:]
|
|
refs = ""
|
|
|
|
# Add the reference of every component in the group and keep a reference
|
|
# to the component so that the other data can be filled in once per group
|
|
for component in group:
|
|
if len(refs) > 0:
|
|
refs += ","
|
|
refs += component.getRef()
|
|
c = component
|
|
|
|
# Fill in the component groups common data
|
|
# columns = ['Item', 'Qty', 'Reference(s)', 'DNI', 'MFRPN', 'Description', 'Manufacturer',
|
|
# 'Lib', 'Part Name', 'Footprint', 'Datasheet'] + sorted(list(columnset))
|
|
|
|
item += 1
|
|
row.append( item )
|
|
row.append( len(group) )
|
|
row.append( refs );
|
|
row.append( c.getField("DNI") ) # DNI
|
|
row.append( c.getField("MFRPN") ) # Mfr's Part #
|
|
row.append( c.getDescription() )
|
|
row.append( c.getField("Manufacturer") )
|
|
row.append( c.getLibName() )
|
|
row.append( c.getPartName() )
|
|
row.append( c.getFootprint() )
|
|
row.append( c.getDocs() )
|
|
|
|
# from column 11 upwards, use the fieldnames to grab the data
|
|
for field in columns[11:]:
|
|
row.append( net.getGroupField(group, field) );
|
|
|
|
writerow( out, row )
|
|
|
|
f.close()
|
|
|