gh-111178: fix UBSan failures in Modules/_ssl.c
#130719
Conversation
|
Probably not the best place to discuss, but I just realized something while applying labels: we probably should have been backporting some of these, otherwise other changes that get backported to these modules will conflict. WDYT? (It's good that Petr is tagged on this, I'd like to hear his opinion too.) |
|
I don't think we should be backporting all of them preemptively. But if you get conflicts, it's probably best to backport the UB PR first. |
|
I haven't backported the changes which caused conflicts on a recent PR but I resolved them manually. I'm afraid that the backports of UB PRs would actually be harder because #128447 was not backported. |
| @@ -153,13 +153,13 @@ _x509name_print(_sslmodulestate *state, X509_NAME *name, int indent, unsigned lo | |||
| * PySSLCertificate_Type | |||
| */ | |||
|
|
|||
| #define _PySSLCertificate_CAST(op) ((PySSLCertificate *)(op)) | |||
There was a problem hiding this comment.
I don't thinkt that this change is useful.
There was a problem hiding this comment.
I think I myself did the change and I at that time put an underscore. But considering I used the convention "objectname+_CAST" for the series of PR and that I amended some changes I've done recently I think we can refactor this one as well (if the code is however not by me and/or is older, I can revert that change)
There was a problem hiding this comment.
Note: it's something I've changed last month, and I thought I fixed all the UBSan issues in the SSL module but this wasn't the case. I can revert it if you don't want this additional diff.
Uh oh!
There was an error while loading. Please reload this page.