Recover from segfault in C code

[Peter Kleiweg]

Is there a way to recover from a segmentation fault triggered by embedded/linked C code? Using a deferred call to recover() doesn't work.

[Ian Lance Taylor]

On Tue, Sep 29, 2015 at 3:25 PM, Peter Kleiweg <pkle...@xs4all.nl> wrote:
> Is there a way to recover from a segmentation fault triggered by embedded/linked C code? Using a deferred call to recover() doesn't work.

There is not. We don't currently support unwinding the stack from a C
function. I think it might be possible in principle--we would have to
discard the current call on the system stack--but it's definitely
tricky.

Ian

[Klaus Post]

On Wednesday, 30 September 2015 00:25:46 UTC+2, Peter Kleiweg wrote:

Is there a way to recover from a segmentation fault triggered by embedded/linked C code? Using a deferred call to recover() doesn't work.

I know you are an experienced developer, so I assume you know that segfaults are much worse in C, and in general shouldn't be recovered, since you likely have 'dirty' memory. Depending on the error there is a large probability the code has overwritten portions of your memory before it hit a memory barrier that triggered the segfault.

Is it impossible for you to fix the segfault in the first place?

/Klaus

[Egon]

I would probably run such C code in a separate process and monitor it. (Obviously, if possible and no high perf needed etc.)

[dave.t...@gmail.com]

    It might be better to take the more general approach of providing a safe setjmp() / longjmp() implementation to C code - that might be easier than trying to unwind, could be used for this purpose, and probably many others as well.  (Might it even work as-is if the goroutine is locked to the OS thread?   Or do the usual implementations of those functions attempt to parse the stack instead of just treating it like a blob?)