On Sat, Feb 17, 2018 at 11:18:26AM -0800, Josh Triplett wrote: > On Sat, Feb 17, 2018 at 05:58:39PM +0100, Luc Van Oostenryck wrote: > > Currently, sparse doesn't handle yet VLA's sizeof(). The size > > of a VLA is considered as zero and the result of sizeof() on > > a VLA is treated as an error with a value of -1. > > > > That's a problem with the check done by the sparse tool, which > > warn when operations like memset() are done with a static count > > which is above some limit. Of course, all this is done with > > unsigned numbers and the -1 from sizeof(VLA) is then considered > > as off-limit. > > > > Sure, size of VLAs should be supported but it's longer term. > > One short term solution would be to do the check with signed > > numbers but that would eat the upper bit of the limit which > > may well be the bound that we would like to check. > > > > So, check instead for sizes of -1, which must come from some > > previous errors that must have already been reported, and do > > not issue the memset() warning in this case. > > I'm concerned about generically ignoring this warning for *all* -1 > sizes, because -1 seems like a very common value to slip through for > other reasons. Losing those very real warnings just to avoid getting a > second warning on a VLA doesn't seem worth it. Yes, I think you're right. > (Also, at the very *least* this would need a comment explaining *why* it > ignores -1.) > > Could you somehow propagate a taintedness on that value that causes the > memset to ignore it? Or just change the dummy error value to 0? Yes, that could be a solution (but changing it to zero seems to me as bad). I'll look for something but I think I'll just better spend a bit time on having real support for VLA's size. Thanks for the review -- Luc -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
