This patch renumbers the metadata nodes in debug info testcases after https://reviews.llvm.org/D26769. This is a separate patch because it causes so much churn. This was implemented with a python script that pipes the testcases through llvm-as - | llvm-dis - and then goes through the original and new output side-by side to insert all comments at a close-enough location.
Here's the script for reference:
import re, sys, subprocess
fn = sys.argv[1]
stdout = subprocess.check_output('llvm-as %s -o - | llvm-dis -o -' % fn, shell=True)
new = stdout.split('\n')
out = []
comment = re.compile(r'^([^;]*)(;.*)$')
ws = re.compile(r'^ *$')
empty = re.compile(r'^ *\n$')
for line in open(fn):
if new and new[0].startswith('; ModuleID ='):
new = new[1:]
# If there was a comment on this line, carry it over.
m = comment.match(line)
if m:
if ws.match(m.group(1)):
if m.group(2).startswith('; Function Attrs'):
continue
out.append(line)
else:
cmnt = m.group(2)
if cmnt.startswith('; preds =') or cmnt.count(r'#uses='):
cmnt = ''
out.append(new[0]+' '+cmnt+'\n')
new = new[1:]
else:
if empty.match(line):
if out[-1] <> '\n':
out.append('\n')
continue
# Try to sync the original and the new output again by comparing the
# first couple of characters.
if new:
if new[0] == '' and out and out[-1] == '\n':
new = new[1:]
continue
out.append(new[0]+'\n')
new = new[1:]
if len(new) > 2 and new[2][:8] == line[:8]:
out.append(new[0]+'\n')
new = new[1:]
if len(new) > 1 and new[1][:8] == line[:8]:
out.append(new[0]+'\n')
new = new[1:]
if len(new) > 0 and new[0][:8] == line[:8]:
out.append(new[0]+'\n')
new = new[1:]
f = open(fn, 'w')
f.write(''.join(out+new)+'\n')
Is dropping all this debug info intentional?
& theer are other 'interesting' changes in other files, but rather than discussing all of them:
Perhaps it'd be better to just run this script over the files that are changed by your previous commit? Ah, I guess that's what you're already doing.
*ponders*