Google Code Exporter

``` After looking at my Ubuntu 10.04 machine again, it looks like I had misread the process map originally. RHEL5 was the only place where the issue actually occurred. By...

``` Nice debugging! I wonder which of these issues are necessary to reproduce the problem in theory, and which just happened to be necessary to trigger it in your case....

``` It looks like the behavior still exists in current kernels (tested with 3.0.0), but only commonly occurs under specific conditions. The kernel has two different strategies for finding unused...

``` Thanks for all the research! It makes sense that topdown vs bottomup would affect how likely this is to occur, since the address space in question is pretty low...

``` While tricks such as /proc/$$/maps (checked once at startup) might work, the cross-platform support seems like more hassle than it's worth. One possibility is for the user to supply...

``` Looking at this, in preparation for a new release (hopefully next week!) I'm wondering if it's not a simpler fix to just replace data_start with _etext. It looks like...

``` I checked with a linker expert, who said __etext should do what we want, so I'm going with that. ``` Original comment by `[email protected]` on 19 Dec 2011 at...

``` That's good to hear. When I tested it, __etext did what we want it to do (especially considering that the function is inlined in a header file.) However, what...

``` Both __etext and _start "should" point to the main executable. I think that is what we want in this case. My expectation is that most uses of TemplateString will...

``` I attached some test cases involving various combinations of _start/__etext referenced by the main executable, a shared library, and the same shared library loaded by dlopen. The case of...