As found in #22931, if an invalid cookie value is found while parsing, the rest of the cookie is silently ignored. The expected behavior is undefined in RFC 6265, but does state that if unexpected values are encountered that user agents MAY ignore an entire set-cookie string (not just a subsection of it). Given that, invalid cookie portions should likely be handled by either:

1. Ignore the cookie string in its entirety and log an error message, or
2. Ignore invalid cookie-pairs but still parse the rest of the string

The latter would likely be the best path (Postel's law and all)

Well, remember that we recently made parsing stricter in response to a security issue...

Sure, but this is in regards to the initial matching, not the parsing. Because the pattern expects RFC conformity, in this cookie string:

Cookie: a=b; c=[; d=r; f=h

The only matching parts will be:

Cookie: a=b;

The rest will be discarded. What I'm proposing is that this should instead result in:

Cookie: a=b; d=r; f=h

The recent changes would stay intact.

Sorry, bad example. Replace "[" in the previous example with any actually invalid character.

The patch at Issue 25228 should partially do what Demian proposed. Anyway, I think Issue 17340 is basically about the same problem.

nosy: + martin.panter
messages: + msg259819

superseder: http.cookies: Handle malformed cookie
resolution: duplicate