skia2/experimental/tools/pdf-comparison.py
Hal Canary a7914a8726 experimental/tools/pdf-comparison: delete worktree when it gets into a bad state
No-Try: true
Change-Id: I61f48fd32348d44a2c7e27d56e404f7876bed3ef
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/245161
Reviewed-by: Hal Canary <halcanary@google.com>
Commit-Queue: Hal Canary <halcanary@google.com>
2019-10-01 14:07:54 +00:00

349 lines
11 KiB
Python
Executable File

#!/usr/bin/env python
# Copyright 2019 Google LLC.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
'''
This tool compares the PDF output of Skia's DM tool of two commits.
It relies on pdfium_test being in the PATH. To build:
mkdir -p ~/src/pdfium
cd ~/src/pdfium
gclient config --unmanaged https://pdfium.googlesource.com/pdfium.git
gclient sync
cd pdfium
gn gen out/default --args='pdf_enable_xfa=false pdf_enable_v8=false pdf_is_standalone=true'
ninja -C out/default pdfium_test
cp out/default/pdfium_test ~/bin/
'''
import os
import re
import shutil
import subprocess
import sys
import tempfile
import threading
EXTRA_GN_ARGS = os.environ.get('PDF_COMPARISON_GN_ARGS', '')
REFERENCE_BACKEND = 'gl' if 'PDF_COMPARISON_NOGPU' not in os.environ else '8888'
DPI = float(os.environ.get('PDF_COMPARISON_DPI', 72))
PDF_CONFIG = 'pdf' if 'PDF_COMPARISON_300DPI' not in os.environ else 'pdf300'
BAD_TESTS = [
'image-cacherator-from-picture',
'image-cacherator-from-raster',
'mixershader',
'shadermaskfilter_image',
'tilemode_decal',
]
NINJA = 'ninja'
PDFIUM_TEST = 'pdfium_test'
NUM_THREADS = int(os.environ.get('PDF_COMPARISON_THREADS', 40))
SOURCES = ['gm']
def test_exe(cmd):
with open(os.devnull, 'w') as o:
try:
subprocess.call([cmd], stdout=o, stderr=o)
except OSError:
return False
return True
def print_cmd(cmd, o):
m = re.compile('[^A-Za-z0-9_./-]')
o.write('+ ')
for c in cmd:
if m.search(c) is not None:
o.write(repr(c) + ' ')
else:
o.write(c + ' ')
o.write('\n')
o.flush()
def check_call(cmd, **kwargs):
print_cmd(cmd, sys.stdout)
return subprocess.check_call(cmd, **kwargs)
def check_output(cmd, **kwargs):
print_cmd(cmd, sys.stdout)
return subprocess.check_output(cmd, **kwargs)
def remove(*paths):
for path in paths:
os.remove(path)
def timeout(deadline, cmd):
#print_cmd(cmd, sys.stdout)
with open(os.devnull, 'w') as o:
proc = subprocess.Popen(cmd, stdout=o, stderr=subprocess.STDOUT)
timer = threading.Timer(deadline, proc.terminate)
timer.start()
proc.wait()
timer.cancel()
return proc.returncode
def is_same(path1, path2):
if not os.path.isfile(path1) or not os.path.isfile(path2):
return os.path.isfile(path1) == os.path.isfile(path2)
with open(path1, 'rb') as f1:
with open(path2, 'rb') as f2:
while True:
c1, c2 = f1.read(4096), f2.read(4096)
if c1 != c2:
return False
if not c1:
return True
def getfilesoftype(directory, ending):
for dirpath, _, filenames in os.walk(directory):
rp = os.path.normpath(os.path.relpath(dirpath, directory))
for f in filenames:
if f.endswith(ending):
yield os.path.join(rp, f)
def get_common_paths(dirs, ext):
return sorted(list(
set.intersection(*(set(getfilesoftype(d, ext)) for d in dirs))))
def printable_path(d):
if 'TMPDIR' in os.environ:
return d.replace(os.path.normpath(os.environ['TMPDIR']) + '/', '$TMPDIR/')
return d
def spawn(cmd):
with open(os.devnull, 'w') as o:
subprocess.Popen(cmd, stdout=o, stderr=o)
def sysopen(arg):
plat = sys.platform
if plat.startswith('darwin'):
spawn(["open", arg])
elif plat.startswith('win'):
# pylint: disable=no-member
os.startfile(arg)
else:
spawn(["xdg-open", arg])
HTML_HEAD = '''
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>DIFF</title>
<style>
body{
background-size:16px 16px;
background-color:rgb(230,230,230);
background-image:
linear-gradient(45deg,rgba(255,255,255,.2) 25%,transparent 25%,transparent 50%,
rgba(255,255,255,.2) 50%,rgba(255,255,255,.2) 75%,transparent 75%,transparent)}
div.r{position:relative;left:0;top:0}
table{table-layout:fixed;width:100%}
img.s{max-width:100%;max-height:320;left:0;top:0}
img.b{position:absolute;mix-blend-mode:difference}
</style>
<script>
function r(c,e,n,g){
t=document.getElementById("t");
function ce(t){return document.createElement(t);}
function ct(n){return document.createTextNode(n);}
function ac(u,v){u.appendChild(v);}
function cn(u,v){u.className=v;}
function it(s){ td=ce("td"); a=ce("a"); a.href=s; img=ce("img"); img.src=s;
cn(img,"s"); ac(a,img); ac(td,a); return td; }
tr=ce("tr"); td=ce("td"); td.colSpan="4"; ac(td, ct(n)); ac(tr,td);
ac(t,tr); tr=ce("tr"); td=ce("td"); dv=ce("div"); cn(dv,"r");
img=ce("img"); img.src=c; cn(img,"s"); ac(dv,img); img=ce("img");
img.src=e; cn(img,"s b"); ac(dv,img); ac(td,dv); ac(tr,td);
ac(tr,it(c)); ac(tr,it(e)); ac(tr,it(g)); ac(t,tr); }
document.addEventListener('DOMContentLoaded',function(){
'''
HTML_TAIL = '''];
for(i=0;i<z.length;i++){
r(c+z[i][0],e+z[i][0],z[i][2],c+z[i][1]);}},false);
</script></head><body><table id="t">
<tr><th>BEFORE-AFTER DIFF</th>
<th>BEFORE</th><th>AFTER</th>
<th>REFERENCE</th></tr>
</table></body></html>'''
def shard(fn, arglist):
jobs = [[arg for j, arg in enumerate(arglist) if j % NUM_THREADS == i]
for i in range(NUM_THREADS)]
results = []
def do_shard(*args):
for arg in args:
results.append(fn(arg))
thread_list = []
for job in jobs:
t = threading.Thread(target=do_shard, args=job)
t.start()
thread_list += [t]
for t in thread_list:
t.join()
return results
def shardsum(fn, arglist):
'return the number of True results returned by fn(arg) for arg in arglist.'
return sum(1 for result in shard(fn, arglist) if result)
def checkout_worktree(checkoutable):
directory = os.path.join(tempfile.gettempdir(), 'skpdf_control_tree')
commit = check_output(['git', 'rev-parse', checkoutable]).strip()
if os.path.isdir(directory):
try:
check_call(['git', 'checkout', commit], cwd=directory)
return directory
except subprocess.CalledProcessError:
shutil.rmtree(directory)
check_call(['git', 'worktree', 'add', '-f', directory, commit])
return directory
def build_skia(directory, executable):
args = ('--args=is_debug=false'
' extra_cflags=["-DSK_PDF_LESS_COMPRESSION",'
' "-DSK_PDF_BASE85_BINARY"] ')
if test_exe('ccache'):
args += ' cc_wrapper="ccache"'
args += EXTRA_GN_ARGS
build_dir = directory + '/out/pdftest'
check_call([sys.executable, 'bin/sync'], cwd=directory)
check_call([directory + '/bin/gn', 'gen', 'out/pdftest', args],
cwd=directory)
check_call([NINJA, executable], cwd=build_dir)
return os.path.join(build_dir, executable)
def build_and_run_dm(directory, data_dir):
dm = build_skia(directory, 'dm')
for source in SOURCES:
os.makedirs(os.path.join(data_dir, PDF_CONFIG, source))
dm_args = [dm, '--src'] + SOURCES + ['--config', PDF_CONFIG, '-w', data_dir]
if BAD_TESTS:
dm_args += ['-m'] + ['~^%s$' % x for x in BAD_TESTS]
check_call(dm_args, cwd=directory)
return dm
def rasterize(path):
ret = timeout(30, [PDFIUM_TEST, '--png', '--scale=%g' % (DPI / 72.0), path])
if ret != 0:
sys.stdout.write(
'\nTIMEOUT OR ERROR [%d] "%s"\n' % (ret, printable_path(path)))
return
assert os.path.isfile(path + '.0.png')
def main(control_commitish):
assert os.pardir == '..' and '/' in [os.sep, os.altsep]
assert test_exe(NINJA)
assert test_exe(PDFIUM_TEST)
os.chdir(os.path.dirname(__file__) + '/../..')
control_worktree = checkout_worktree(control_commitish)
tmpdir = tempfile.mkdtemp(prefix='skpdf_')
exp = tmpdir + '/experim'
con = tmpdir + '/control'
build_and_run_dm(os.curdir, exp)
dm = build_and_run_dm(control_worktree, con)
image_diff_metric = build_skia(control_worktree, 'image_diff_metric')
out = sys.stdout
common_paths = get_common_paths([con, exp], '.pdf')
out.write('\nNumber of PDFs: %d\n\n' % len(common_paths))
def compare_identical(path):
cpath, epath = (os.path.join(x, path) for x in (con, exp))
if is_same(cpath, epath):
remove(cpath, epath)
return True
return False
identical_count = shardsum(compare_identical, common_paths)
out.write('Number of identical PDFs: %d\n\n' % identical_count)
differing_paths = get_common_paths([con, exp], '.pdf')
if not differing_paths:
out.write('All PDFs are the same!\n')
sys.exit(0)
out.write('Number of differing PDFs: %d\n' % len(differing_paths))
for p in differing_paths:
out.write(' %s\n' % printable_path(tmpdir + '/*/' + p))
out.write('\n')
shard(rasterize,
[os.path.join(x, p) for p in differing_paths for x in [con, exp]])
common_pngs = get_common_paths([con, exp], '.pdf.0.png')
identical_count = shardsum(compare_identical, common_pngs)
out.write('Number of PDFs that rasterize the same: %d\n\n'
% identical_count)
differing_pngs = get_common_paths([con, exp], '.pdf.0.png')
if not differing_pngs:
out.write('All PDFs rasterize the same!\n')
sys.exit(0)
out.write('Number of PDFs that rasterize differently: %d\n'
% len(differing_pngs))
for p in differing_pngs:
out.write(' %s\n' % printable_path(tmpdir + '/*/' + p))
out.write('\n')
scores = dict()
def compare_differing_pngs(path):
cpath, epath = (os.path.join(x, path) for x in (con, exp))
s = float(subprocess.check_output([image_diff_metric, cpath, epath]))
indicator = '.' if s < 0.001 else ':' if s < 0.01 else '!'
sys.stdout.write(indicator)
sys.stdout.flush()
scores[path] = s
shard(compare_differing_pngs, differing_pngs)
paths = sorted(scores.iterkeys(), key=lambda p: -scores[p])
out.write('\n\n')
for p in paths:
pdfpath = printable_path(tmpdir + '/*/' + p.replace('.0.png', ''))
out.write(' %6.4f %s\n' % (scores[p], pdfpath))
out.write('\n')
errors = []
rc = re.compile('^' + PDF_CONFIG + r'/([^/]*)/([^/]*)\.pdf\.0\.png$')
for p in paths:
m = rc.match(p)
assert(m)
source, name = m.groups()
errors.append((source, name, scores[p]))
for source in SOURCES:
os.makedirs(os.path.join(con, REFERENCE_BACKEND, source))
dm_args = [dm, '--src'] + SOURCES + [
'--config', REFERENCE_BACKEND, '-w', con, '-m'] + [
'^%s$' % name for _, name, _ in errors]
check_call(dm_args, cwd=control_worktree)
report = tmpdir + '/report.html'
with open(report, 'w') as o:
o.write(HTML_HEAD)
o.write('c="%s/";\n' % os.path.relpath(con, tmpdir))
o.write('e="%s/";\n' % os.path.relpath(exp, tmpdir))
o.write('z=[\n')
for source, name, score in errors:
gt = REFERENCE_BACKEND + '/' + source + '/' + name + '.png'
p = '%s/%s/%s.pdf.0.png' % (PDF_CONFIG, source, name)
desc = '%s | %s | %g' % (source, name, score)
o.write('["%s","%s","%s"],\n' % (p, gt, desc))
o.write(HTML_TAIL)
out.write(printable_path(report) + '\n')
sysopen(report)
if __name__ == '__main__':
if len(sys.argv) != 2:
USAGE = ('\nusage:\n {0} COMMIT_OR_BRANCH_TO_COMPARE_TO\n\n'
'e.g.:\n {0} HEAD\nor\n {0} HEAD~1\n\n')
sys.stderr.write(USAGE.format(sys.argv[0]))
sys.exit(1)
main(sys.argv[1])