| 1 |
sanchan wrote: |
| 2 |
> missing -fno-strict-aliasing wherever there are bad programming |
| 3 |
> practices and so on; |
| 4 |
|
| 5 |
IIUC, GCC's current -Wstrict-aliasing implementation has a number of |
| 6 |
flaws. According to a recent thread[i] on the GCC ml.. |
| 7 |
|
| 8 |
> The current implementation of -Wstrict-aliasing looks at a single expression |
| 9 |
> at once. Although it does OK given the limited scope, it has several drawbacks. |
| 10 |
> First, it produces false positives, i.e., it warns when it should not. |
| 11 |
> For instance, it warns about pointer conversions even when the pointers |
| 12 |
> are never dereferenced. |
| 13 |
> Second, it has many false negatives, i.e., it does not warn when it should. |
| 14 |
> It does not warn when an address is not taken, for instance: |
| 15 |
> float *f = ... |
| 16 |
> int *i = (int*)f; |
| 17 |
> Third, it does not display the name or type of the aliased objects, which |
| 18 |
> makes the warnings hard to understand in templates or macros. |
| 19 |
|
| 20 |
Until that improves it might be a good idea to take strict aliasing |
| 21 |
warnings with a grain of salt. |
| 22 |
|
| 23 |
|
| 24 |
[i] http://permalink.gmane.org/gmane.comp.gcc.patches/131449 |
| 25 |
|
| 26 |
-- |
| 27 |
by design, by neglect |
| 28 |
dirtyepic gentoo org for a fact or just for effect |
| 29 |
9B81 6C9F E791 83BB 3AB3 5B2D E625 A073 8379 37E8 (0x837937E8) |
Archives