On 9/18/18 5:44 AM, Bernd Edlinger wrote:
>>
>
> Hmm, you know, I wrote a while ago the following:
> https://gcc.gnu.org/ml/gcc-patches/2018-09/msg00411.html
>
> Where I suggested to change c_strlen parameter nonstr to a
> structure, where additional information could go.
It certainly seems better than continually adding more parameters and/or
having a given parameter have multiple overloaded meanings. We'd fill
in everything we can and let the caller determine what bits they need.
Martin has suggested something similar.
>
> Well, I have the impression that the 6/6 patch should be split
> up in a part that uses c_strlen in its current form, and a
> follow-up patch that adds any additional info.
> I prefer doing things step-by-step, you know.
I'm pretty sure it's not going to be able to use c_strlen in its current
form. There are times when it really wants the length of the string,
even if it's not terminated.
>
> Note that the example given is probably unsafe:
>> T (&b[3][1] + v0, bsz);
>
> This does not pass the check in string_constant because it
> looks like _2 = (char*)&b + _1; _3 = strlen (_2);
> I have no way to prove that v0 is not larger than sizeof (b[3]).
That's OK. The existence of the variable part turns "must" warnings
into "may" warnings precisely because we don't know how v0 impacts things.
>
> By the way test test case for pr87053 is an example
> of a false positive warning (it comes twice even though
> TREE_NO_WARNING is used).
>
> gcc pr87053.c
> pr87053.c: In function 'main':
> pr87053.c:15:26: warning: 'strlen' argument missing terminating nul
> [-Wstringop-overflow=]
> 15 | if (__builtin_strlen (u.z) != 7)
> | ~^~
> pr87053.c:11:3: note: referenced argument declared here
> 11 | } u = {{"1234", "567"}};
> | ^
> pr87053.c:15:26: warning: 'strlen' argument missing terminating nul
> [-Wstringop-overflow=]
> 15 | if (__builtin_strlen (u.z) != 7)
> | ~^~
> pr87053.c:11:3: note: referenced argument declared here
> 11 | } u = {{"1234", "567"}};
> | ^
>
> I think the duplicate warnings will not go away unless
> this warning is only diagnosed in the tree-ssa-strlen
> pass instead of the folding code, which happens repeatedly.
Please file a bug. FWIW, I wouldn't necessarily rely on on the strlen
pass not being repeated either.
jeff
