mirror of
https://github.com/DJSundog/NopSCADlib.git
synced 2024-11-30 10:40:03 -05:00
407 lines
19 KiB
Python
Executable File
407 lines
19 KiB
Python
Executable File
#!/usr/bin/env python
|
|
|
|
#
|
|
# NopSCADlib Copyright Chris Palmer 2018
|
|
# nop.head@gmail.com
|
|
# hydraraptor.blogspot.com
|
|
#
|
|
# This file is part of NopSCADlib.
|
|
#
|
|
# NopSCADlib 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 3 of
|
|
# the License, or (at your option) any later version.
|
|
#
|
|
# NopSCADlib is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
|
|
# without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
|
|
# See the GNU General Public License for more details.
|
|
#
|
|
# You should have received a copy of the GNU General Public License along with NopSCADlib.
|
|
# If not, see <https://www.gnu.org/licenses/>.
|
|
#
|
|
|
|
#
|
|
#! Generates exploded and unexploded assembly views and scrapes build instructions to make readme.md, readme.html and printme.html files for the project.
|
|
#
|
|
from __future__ import print_function
|
|
from set_config import *
|
|
import openscad
|
|
from tests import do_cmd, update_image, colour_scheme, background
|
|
import time
|
|
import times
|
|
import options
|
|
from deps import *
|
|
import os
|
|
import json
|
|
import blurb
|
|
import bom
|
|
import shutil
|
|
from colorama import Fore
|
|
|
|
def is_assembly(s):
|
|
return s[-9:] == '_assembly' or s[-11:] == '_assemblies'
|
|
|
|
def bom_to_assemblies(bom_dir, bounds_map):
|
|
global flat_bom
|
|
#
|
|
# Make a list of all the parts in the BOM
|
|
#
|
|
bom_file = bom_dir + "/bom.json"
|
|
with open(bom_file) as json_file:
|
|
flat_bom = json.load(json_file)
|
|
#
|
|
# Decide if we need big or small assembly pictures
|
|
#
|
|
for bom in flat_bom:
|
|
if bom["big"] == None:
|
|
big = False
|
|
for ass in bom["assemblies"]:
|
|
for b in flat_bom:
|
|
if b["name"] == ass:
|
|
if b["big"]:
|
|
big = True
|
|
break
|
|
if not big:
|
|
for stl in bom["printed"]:
|
|
bounds = bounds_map[stl]
|
|
width = bounds[1][0] - bounds[0][0]
|
|
depth = bounds[1][1] - bounds[0][1]
|
|
if max(width, depth) > 80:
|
|
big = True
|
|
break
|
|
bom["big"] = big or bom["routed"]
|
|
#
|
|
# Remove the main assembly if it is a shell
|
|
#
|
|
if flat_bom:
|
|
ass = flat_bom[-1]
|
|
if len(ass["assemblies"]) < 2 and not ass["vitamins"] and not ass["printed"] and not ass["routed"]:
|
|
flat_bom = flat_bom[:-1]
|
|
return [assembly["name"] for assembly in flat_bom]
|
|
|
|
def eop(print_mode, doc_file, last = False, first = False):
|
|
if print_mode:
|
|
print('\n<div style="page-break-after: always;"></div>', file = doc_file)
|
|
else:
|
|
if not first:
|
|
print('[Top](#TOP)', file = doc_file)
|
|
if not last:
|
|
print("\n---", file = doc_file)
|
|
|
|
def pad(s, before, after = 0):
|
|
return ' ' * before + str(s) + ' ' * after
|
|
|
|
def titalise(name):
|
|
cap_next = True
|
|
result = ''
|
|
for c in name.replace('_', ' '):
|
|
result = result + (c.upper() if cap_next else c);
|
|
cap_next = c == ' '
|
|
return result
|
|
|
|
def usage():
|
|
print("\nusage:\n\t views [target_config] [<name1>_assembly] ... [<nameN>_assembly] - Create assembly images and readme.")
|
|
sys.exit(1)
|
|
|
|
def views(target, do_assemblies = None):
|
|
done_assemblies = []
|
|
#
|
|
# Make the target directory
|
|
#
|
|
top_dir = set_config(target, usage)
|
|
target_dir = top_dir + 'assemblies'
|
|
deps_dir = top_dir + "deps"
|
|
bom_dir = top_dir + "bom"
|
|
if not os.path.isdir(target_dir):
|
|
os.makedirs(target_dir)
|
|
if not os.path.isdir(deps_dir):
|
|
os.makedirs(deps_dir)
|
|
|
|
times.read_times(target_dir)
|
|
options.check_options(deps_dir)
|
|
bounds_fname = top_dir + 'stls/bounds.json'
|
|
with open(bounds_fname) as json_file:
|
|
bounds_map = json.load(json_file)
|
|
#
|
|
# Find all the assemblies and remove any old views
|
|
#
|
|
assemblies = bom_to_assemblies(bom_dir, bounds_map)
|
|
lc_assemblies = [ass.lower() for ass in assemblies]
|
|
for file in os.listdir(target_dir):
|
|
if file.endswith('.png'):
|
|
assembly = file[:-4].replace('_assembled', '_assembly')
|
|
if assembly.endswith('_tn'):
|
|
assembly = assembly[:-3]
|
|
if not assembly in assemblies:
|
|
print("Removing %s" % file)
|
|
os.remove(target_dir + '/' + file)
|
|
#
|
|
# Find all the scad files
|
|
#
|
|
main_blurb = None
|
|
for dir in source_dirs(bom_dir):
|
|
if os.path.isdir(dir):
|
|
for filename in os.listdir(dir):
|
|
if filename.endswith('.scad'):
|
|
#
|
|
# find any modules with names ending in _assembly
|
|
#
|
|
with open(dir + "/" + filename, "r") as f:
|
|
lines = f.readlines()
|
|
line_no = 0
|
|
for line in lines:
|
|
words = line.split()
|
|
if len(words) and words[0] == "module":
|
|
module = words[1].split('(')[0]
|
|
if is_assembly(module):
|
|
lc_module = module.lower()
|
|
if lc_module in lc_assemblies:
|
|
real_name = assemblies[lc_assemblies.index(lc_module)]
|
|
#
|
|
# Scrape the assembly instructions
|
|
#
|
|
for ass in flat_bom:
|
|
if ass["name"] == real_name:
|
|
if not "blurb" in ass:
|
|
ass["blurb"] = blurb.scrape_module_blurb(lines[:line_no])
|
|
break
|
|
if not do_assemblies or real_name in do_assemblies:
|
|
#
|
|
# Run openscad on the created file
|
|
#
|
|
dname = deps_name(deps_dir, filename)
|
|
for explode in [0, 1]:
|
|
png_name = target_dir + '/' + real_name + '.png'
|
|
if not explode:
|
|
png_name = png_name.replace('_assembly', '_assembled')
|
|
changed = check_deps(png_name, dname)
|
|
changed = times.check_have_time(changed, png_name)
|
|
changed = options.have_changed(changed, png_name)
|
|
tmp_name = 'tmp.png'
|
|
if changed:
|
|
print(changed)
|
|
#
|
|
# make a file to use the module
|
|
#
|
|
png_maker_name = 'png.scad'
|
|
with open(png_maker_name, "w") as f:
|
|
f.write("use <%s/%s>\n" % (dir, filename))
|
|
f.write("%s();\n" % module);
|
|
t = time.time()
|
|
openscad.run_list(options.list() + ["-D$pose=1", "-D$explode=%d" % explode, colour_scheme, "--projection=p", "--imgsize=4096,4096", "--autocenter", "--viewall", "-d", dname, "-o", tmp_name, png_maker_name]);
|
|
times.add_time(png_name, t)
|
|
do_cmd(["magick", tmp_name, "-trim", "-resize", "1004x1004", "-bordercolor", background, "-border", "10", tmp_name])
|
|
update_image(tmp_name, png_name)
|
|
os.remove(png_maker_name)
|
|
tn_name = png_name.replace('.png', '_tn.png')
|
|
if mtime(png_name) > mtime(tn_name):
|
|
do_cmd(("magick "+ png_name + " -trim -resize 280x280 -background " + background + " -gravity Center -extent 280x280 -bordercolor " + background + " -border 10 " + tmp_name).split())
|
|
update_image(tmp_name, tn_name)
|
|
done_assemblies.append(real_name)
|
|
else:
|
|
if module == 'main_assembly':
|
|
main_blurb = blurb.scrape_module_blurb(lines[:line_no])
|
|
line_no += 1
|
|
times.print_times()
|
|
#
|
|
# Build the document
|
|
#
|
|
for print_mode in [True, False]:
|
|
doc_name = top_dir + "readme.md"
|
|
with open(doc_name, "wt") as doc_file:
|
|
#
|
|
# Title, description and picture
|
|
#
|
|
project = ' '.join(word[0].upper() + word[1:] for word in os.path.basename(os.getcwd()).split('_'))
|
|
print('<a name="TOP"></a>\n# %s' % project, file = doc_file)
|
|
main_file = bom.find_scad_file('main_assembly')
|
|
if not main_file:
|
|
raise Exception("can't find source for main_assembly")
|
|
text = blurb.scrape_blurb(source_dir + '/' + main_file)
|
|
if len(text):
|
|
print(text, file = doc_file, end = '')
|
|
else:
|
|
if print_mode:
|
|
print(Fore.MAGENTA + "Missing project description" + Fore.WHITE)
|
|
print('![Main Assembly](assemblies/%s.png)\n' % flat_bom[-1]["name"].replace('_assembly', '_assembled'), file = doc_file)
|
|
eop(print_mode, doc_file, first = True)
|
|
#
|
|
# Build TOC
|
|
#
|
|
print('## Table of Contents', file = doc_file)
|
|
print('1. [Parts list](#Parts_list)', file = doc_file)
|
|
for ass in flat_bom:
|
|
name = ass["name"]
|
|
cap_name = titalise(name)
|
|
print('1. [%s](#%s)' % (cap_name, name), file = doc_file)
|
|
print(file = doc_file)
|
|
eop(print_mode, doc_file)
|
|
#
|
|
# Global BOM
|
|
#
|
|
print('<a name="Parts_list"></a>\n## Parts list', file = doc_file)
|
|
types = ["vitamins", "printed", "routed"]
|
|
headings = {"vitamins" : "vitamins", "printed" : "3D printed parts", "routed" : "CNC routed parts"}
|
|
things = {}
|
|
for t in types:
|
|
things[t] = {}
|
|
for ass in flat_bom:
|
|
for t in types:
|
|
for thing in ass[t]:
|
|
if thing in things[t]:
|
|
things[t][thing] += ass[t][thing]["count"]
|
|
else:
|
|
things[t][thing] = ass[t][thing]["count"]
|
|
for ass in flat_bom:
|
|
name = titalise(ass["name"][:-9]).replace(' ',' ')
|
|
print('| <span style="writing-mode: vertical-rl; text-orientation: mixed;">%s</span> ' % name, file = doc_file, end = '')
|
|
print('| <span style="writing-mode: vertical-rl; text-orientation: mixed;">TOTALS</span> | |', file = doc_file)
|
|
|
|
print(('|---:' * len(flat_bom) + '|---:|:---|'), file = doc_file)
|
|
|
|
for t in types:
|
|
if things[t]:
|
|
totals = {}
|
|
heading = headings[t][0:1].upper() + headings[t][1:]
|
|
print(('| ' * len(flat_bom) + '| | **%s** |') % heading, file = doc_file)
|
|
for thing in sorted(things[t], key = lambda s: s.split(":")[-1]):
|
|
for ass in flat_bom:
|
|
count = ass[t][thing]["count"] if thing in ass[t] else 0
|
|
print('| %s ' % pad(count if count else '.', 2, 1), file = doc_file, end = '')
|
|
name = ass["name"]
|
|
if name in totals:
|
|
totals[name] += count
|
|
else:
|
|
totals[name] = count
|
|
print('| %s | %s |' % (pad(things[t][thing], 2, 1), pad(thing.split(":")[-1], 2)), file = doc_file)
|
|
|
|
grand_total = 0
|
|
for ass in flat_bom:
|
|
name = ass["name"]
|
|
total = totals[name] if name in totals else 0
|
|
print('| %s ' % pad(total if total else '.', 2, 1), file = doc_file, end = '')
|
|
grand_total += total
|
|
print("| %s | %s |" % (pad(grand_total, 2, 1), pad('Total %s count' % headings[t], 2)), file = doc_file)
|
|
|
|
print(file = doc_file)
|
|
eop(print_mode, doc_file)
|
|
#
|
|
# Assembly instructions
|
|
#
|
|
for ass in flat_bom:
|
|
name = ass["name"]
|
|
cap_name = titalise(name)
|
|
|
|
if ass["count"] > 1:
|
|
print('<a name="%s"></a>\n## %d x %s' % (name, ass["count"], cap_name), file = doc_file)
|
|
else:
|
|
print('<a name="%s"></a>\n## %s' % (name, cap_name), file = doc_file)
|
|
vitamins = ass["vitamins"]
|
|
if vitamins:
|
|
print("### Vitamins", file = doc_file)
|
|
print("|Qty|Description|", file = doc_file)
|
|
print("|---:|:----------|", file = doc_file)
|
|
for v in sorted(vitamins, key = lambda s: s.split(":")[-1]):
|
|
print("|%d|%s|" % (vitamins[v]["count"], v.split(":")[1]), file = doc_file)
|
|
print("\n", file = doc_file)
|
|
|
|
printed = ass["printed"]
|
|
if printed:
|
|
print('### 3D Printed parts', file = doc_file)
|
|
keys = sorted(list(printed.keys()))
|
|
for i in range(len(keys)):
|
|
p = keys[i]
|
|
print('%s %d x %s |' % ('\n|' if not (i % 3) else '', printed[p]["count"], p), file = doc_file, end = '')
|
|
if (i % 3) == 2 or i == len(printed) - 1:
|
|
n = (i % 3) + 1
|
|
print('\n|%s' % ('---|' * n), file = doc_file)
|
|
for j in range(n):
|
|
part = keys[i - n + j + 1]
|
|
print('| ![%s](stls/%s) %s' % (part, part.replace('.stl','.png'), '|\n' if j == j - 1 else ''), end = '', file = doc_file)
|
|
print('\n', file = doc_file)
|
|
print('\n', file = doc_file)
|
|
|
|
routed = ass["routed"]
|
|
if routed:
|
|
print("### CNC Routed parts", file = doc_file)
|
|
keys = sorted(list(routed.keys()))
|
|
for i in range(len(keys)):
|
|
r = keys[i]
|
|
print('%s %d x %s |' % ('\n|' if not (i % 3) else '', routed[r]["count"], r), file = doc_file, end = '')
|
|
if (i % 3) == 2 or i == len(routed) - 1:
|
|
n = (i % 3) + 1
|
|
print('\n|%s' % ('---|' * n), file = doc_file)
|
|
for j in range(n):
|
|
part = keys[i - n + j + 1]
|
|
print('| ![%s](dxfs/%s) %s' % (part, part.replace('.dxf','.png'), '|\n' if j == j - 1 else ''), end = '', file = doc_file)
|
|
print('\n', file = doc_file)
|
|
print('\n', file = doc_file)
|
|
|
|
sub_assemblies = ass["assemblies"]
|
|
if sub_assemblies:
|
|
print("### Sub-assemblies", file = doc_file)
|
|
keys = sorted(list(sub_assemblies.keys()))
|
|
for i in range(len(keys)):
|
|
a = keys[i]
|
|
print('%s %d x %s |' % ('\n|' if not (i % 3) else '', sub_assemblies[a], a), file = doc_file, end = '')
|
|
if (i % 3) == 2 or i == len(keys) - 1:
|
|
n = (i % 3) + 1
|
|
print('\n|%s' % ('---|' * n), file = doc_file)
|
|
for j in range(n):
|
|
a = keys[i - n + j + 1].replace('_assembly', '_assembled')
|
|
print('| ![%s](assemblies/%s) %s' % (a, a + '_tn.png', '|\n' if j == j - 1 else ''), end = '', file = doc_file)
|
|
print('\n', file = doc_file)
|
|
print('\n', file = doc_file)
|
|
|
|
small = not ass["big"]
|
|
suffix = '_tn.png' if small else '.png'
|
|
print('### Assembly instructions', file = doc_file)
|
|
print('![%s](assemblies/%s)\n' % (name, name + suffix), file = doc_file)
|
|
|
|
if "blurb" in ass and ass["blurb"]:
|
|
print(ass["blurb"], file = doc_file)
|
|
else:
|
|
if print_mode:
|
|
print(Fore.MAGENTA + "Missing instructions for %s" % name, Fore.WHITE)
|
|
|
|
name = name.replace('_assembly', '_assembled')
|
|
print('![%s](assemblies/%s)\n' % (name, name + suffix), file = doc_file)
|
|
eop(print_mode, doc_file, last = ass == flat_bom[-1] and not main_blurb)
|
|
#
|
|
# If main module is suppressed print any blurb here
|
|
#
|
|
if main_blurb:
|
|
print(main_blurb, file = doc_file)
|
|
eop(print_mode, doc_file, last = True)
|
|
#
|
|
# Convert to HTML
|
|
#
|
|
html_name = "printme.html" if print_mode else "readme.html"
|
|
with open(top_dir + html_name, "wt") as html_file:
|
|
do_cmd(("python -m markdown -x tables -x sane_lists " + doc_name).split(), html_file)
|
|
#
|
|
# Spell check
|
|
#
|
|
do_cmd('codespell -L od readme.md'.split())
|
|
#
|
|
# List the ones we didn't find
|
|
#
|
|
missing = set()
|
|
for assembly in assemblies + (do_assemblies if do_assemblies else []):
|
|
if assembly not in done_assemblies:
|
|
missing.add(assembly)
|
|
if missing:
|
|
for assembly in missing:
|
|
print(Fore.MAGENTA + "Could not find a module called", assembly, Fore.WHITE)
|
|
sys.exit(1)
|
|
|
|
if __name__ == '__main__':
|
|
if len(sys.argv) > 1 and sys.argv[1][-9:] != "_assembly":
|
|
target, assemblies = sys.argv[1], sys.argv[2:]
|
|
else:
|
|
target, assemblies = None, sys.argv[1:]
|
|
|
|
for a in assemblies:
|
|
if a[-9:] != "_assembly": usage()
|
|
|
|
views(target, assemblies)
|