Re: binary search debugging of compilers

On Sun, May 14, 2023 at 22:42 Thomas Koenig <tkoenig@netcologne.de> wrote:


> Russ Cox <rsc@swtch.com> schrieb:
>
> > As I mentioned at the top, I am interested to hear about earlier uses
> > of approaches like this, or any other ideas for situations where the
> > approach might be applicable. Clearly the general problem has overlaps
> > with group testing [1], and in particular Hwang's binary search-based
> > approach (1972) [2].
>
> For GCC regression-hunting, two techniques are routinely used.
>
> One of them is the use of "gcc bisect", described at
> https://git-scm.com/docs/git-bisect
> . If there is a failing test
> case, this can (at the cost of some CPU time) usually pinpoint
> offending commit.




Ah, this is close to my heart. The story I heard was that automated
regression bisection originated in the Cray compiler group around the
mid 1990s. I didn't join that team until the aughts, but can attest to
the quality of Cray's system, which is still in use today AFAIK.


https://ieeexplore.ieee.org/document/625082


For the general case of reducing runtime errors, the Wolf Fencing algorithm
would be a good thread to pull for uncovering the history.


https://dl.acm.org/doi/abs/10.1145/358690.358695


Hope that helps,
Cam

Return to the comp.compilers page.
Search the comp.compilers archives again.