Merge lp:~brian-murray/apport/bug-1016380 into lp:~apport-hackers/apport/trunk

[Brian Murray]

All the bug patterns are strings so log files should be decoded so that the patterns won't crash on them.

[Martin Pitt (pitti)]

I added tests for this bug.

Why do you need to encode the regexp into UTF-8 when you already decode the UTF-8 string? I think the re and the text both have to have the same type.

We also need to handle the case where the bytes object is not UTF-8, otherwise the .decode('UTF-8') will crash with a UnicodeDecodeError.

So I think it's better to use re in bytes mode when the value is in bytes, as this is both more efficient (avoids decoding) more robust (avoids UnicodeDecodeErrors), and also allows us to match against binary values.

Thanks for the initial fix!

review: Needs Fixing

[Martin Pitt (pitti)]

Fixed in http://bazaar.launchpad.net/~apport-hackers/apport/trunk/revision/2482, with tests. Thanks!

Unmerged revisions

2482. By Brian Murray on 2012-09-07

decode byte objects before trying to use a string regex with them

=== modified file 'apport/report.py'
--- apport/report.py	2012-08-31 10:42:24 +0000
+++ apport/report.py	2012-09-07 20:57:23 +0000
@@ -137,7 +137,10 @@
                 regexp = c.childNodes[0].nodeValue
                 v = report[key]
                 if isinstance(v, problem_report.CompressedValue):
-                    v = v.get_value()
+                    v = v.get_value().decode('UTF-8')
+                    regexp = regexp.encode('UTF-8')
+                elif isinstance(v, bytes):
+                    v = v.get_value().decode('UTF-8')
                     regexp = regexp.encode('UTF-8')
                 try:
                     re_c = re.compile(regexp)