418558 - Detect use of memcpy/memset on non-POD types

Status: RESOLVED DUPLICATE of bug 1411029

(Developer Infrastructure :: Source Code Analysis, defect)

Description

[Matthew Gregan [:kinetik]]

While working on bug #418172, it occurred to me that being able to detect and warn about some unsafe operations on non-POD types would be very useful.  This seems like a useful check that our static analysis tools could perform.

Specifically I was thinking about detecting when a non-POD type is used as the source or destination of a memcpy (what happened in bug #418172 was that a class containing an nsCOMPtr was copying using memcpy, causing a ref counting problem), but there are other cases that would be equally useful to detect (e.g. memset, probably others too).

Comment 1

[(dormant account)]

Matthew, 
Dehydra GCC should be pretty good at detecting these already. Let me know if you want some help in drafting up a script for it.

Comment 2

[Matthew Gregan [:kinetik]]

Attachment: first attempt

Comment 3

[David Mandelin [:dmandelin]]

Looks good. Feel free to file a bug for the member access information or the crashes. The GCC ICEs are probably segfaults within the plugin--at least that's what they always were when I was working on my plugin.

Comment 4

[David Mandelin [:dmandelin]]

Oops, looked like Taras already put in the isStatic attributes.

Comment 5

[Matthew Gregan [:kinetik]]

Actually the comment about ICEs in the script is a bit out of date. I was seeing them with an earlier version of the script, but I rewrote the script to be a bit more efficient and haven't seen them since, so I was probably running out of heap with the earlier version.  I've run a complete clobber build with the plugin and my script successfully (although I did have to make two Dehydra asserts non-fatal).

I'll be busy for a few days before the freeze, but I'll file bugs on the member access, asserts, and location info soon afterwards.

Comment 6

[Nathan Froyd [:froydnj]]

I believe this has mostly been accomplished by bug 1411029, via gcc-8 complaining about such memcpys.