summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--pym/repoman/checks.py61
1 files changed, 48 insertions, 13 deletions
diff --git a/pym/repoman/checks.py b/pym/repoman/checks.py
index 7ec670804..de7ceee23 100644
--- a/pym/repoman/checks.py
+++ b/pym/repoman/checks.py
@@ -118,10 +118,15 @@ class EbuildQuote(ContentCheck):
"""Ensure ebuilds have valid quoting around things like D,FILESDIR, etc..."""
repoman_check_name = 'ebuild.minorsyn'
-
- missing_quotes = re.compile(r'[^"]\${?(D|S|T|ROOT|FILESDIR|WORKDIR)}?\W')
- missing_quotes_exclude = re.compile(r'\[\[.*[^"]\${?(D|S|T|ROOT|FILESDIR|WORKDIR)}?\W.*\]\]')
- ignore_line = re.compile(r'(^$)|(^(\t)*#)')
+ ignore_line = re.compile(r'(^$)|(^\s*#.*)')
+ var_names = r'(D|S|T|ROOT|FILESDIR|WORKDIR)'
+ var_reference = re.compile(r'\$({'+var_names+'}|' + \
+ r'\$' + var_names + '\W)')
+ missing_quotes = re.compile(r'(\s|^)[^"\s]*\${?' + var_names + \
+ r'}?[^"\s]*(\s|$)')
+ var_assignment = re.compile(r'^\s*\w*=.*')
+ cond_begin = re.compile(r'(^|\s+)\[\[($|\\$|\s+)')
+ cond_end = re.compile(r'(^|\s+)\]\]($|\\$|\s+)')
def __init__(self, contents):
ContentCheck.__init__(self, contents)
@@ -133,16 +138,46 @@ class EbuildQuote(ContentCheck):
errors = []
for num, line in enumerate(self.contents):
- match = self.ignore_line.match(line)
- if match:
+ if self.ignore_line.match(line) is not None:
+ continue
+ if self.var_reference.search(line) is None:
continue
- missing_quotes_line = self.missing_quotes.search(line)
- if missing_quotes_line:
- for group in missing_quotes_line.group():
- match = self.missing_quotes_exclude.search(group)
- if not match:
- errors.append((num + 1, MISSING_QUOTES_ERROR))
- break
+ # There can be multiple matches / violations on a single line. We
+ # have to make sure none of the matches are violators. Once we've
+ # found one violator, any remaining matches on the same line can
+ # be ignored.
+ pos = 0
+ while pos <= len(line) - 1:
+ missing_quotes = self.missing_quotes.search(line, pos)
+ if not missing_quotes:
+ break
+ # If the last character of the previous match is a whitespace
+ # character, that character may be needed for the next
+ # missing_quotes match, so search overlaps by 1 character.
+ group = missing_quotes.group()
+ pos = missing_quotes.end() - 1
+
+ # Filter out some false positives that can
+ # get through the missing_quotes regex.
+ if self.var_reference.search(group) is None:
+ continue
+ if self.var_assignment.search(group) is not None:
+ continue
+
+ # This is an attempt to avoid false positives without getting
+ # too complex, while possibly allowing some (hopefully
+ # unlikely) violations to slip through. We just assume
+ # everything is correct if the there is a ' [[ ' or a ' ]] '
+ # anywhere in the whole line (possibly continued over one
+ # line).
+ if self.cond_begin.search(line) is not None:
+ continue
+ if self.cond_end.search(line) is not None:
+ continue
+
+ errors.append((num + 1, MISSING_QUOTES_ERROR))
+ # Any remaining matches on the same line can be ignored.
+ break
return errors