On 05/21/2015 08:19 PM, Martin Sebor wrote: > A program I instrumented to help me debug an otherwise unrelated > problem in 5.1.0 has been crashing in calls to > __builtin_return_address. After checking the manual, I didn't > think I was doing anything wrong. I then did some debugging and > found that the function simply isn't safe to call with non-zero > arguments near the top of the stack. That seemed like a bug to > me so I created a small test case and ran it on a few targets > to see if the problem was isolated to just powerpc (where I'm > working at the moment) or more general. It turned out not to > be target-specific. Before opening a bug, I checked Bugzilla > to see if it's already been reported but couldn't find any open > reports. To be sure I wasn't missing something, I expanded my > search to already resolved bugs. That's when I finally found > pr8743 which had been closed years ago as a documentation issue, > after adding the following to the manual: > > This function should only be used with a nonzero argument > for debugging purposes. > > Since I was using the function exactly for this purpose, I'd > like to propose the patch below to clarify the effects of the > function to set the right expectations and help others avoid > the effort it took me to figure out this is by design. > > Does anyone have any concerns with this update or is it okay > to check in?
Sounds like a good candidate for a warning. Thanks, Pedro Alves
