llvm dev - Mar 2015 - [LLVMdev] [asan] The default value of 'allocator_may_return_null'

Hi,

There seems to be a discrepancy between what the default value of
'allocator_may_return_null' should be:
  source code:
  ./lib/sanitizer_common/sanitizer_flags.inc:COMMON_FLAG(bool,
allocator_may_return_null, false,

 docs (https://code.google.com/p/address-sanitizer/wiki/Flags):
allocator_may_return_null 	1 	If false, the allocator will crash instead of
returning 0 on out-of-memory.

I suspect that the reason to have it default to 'false' is because that
would provide for a more predictable out of memory behavior in most use cases.
Is that the case or should this be system dependent and match the default system
allocator?

Thanks,
Anna.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150310/dbfa8441/attachment.html>

On Tue, Mar 10, 2015 at 4:40 PM, Anna Zaks <ganna at apple.com> wrote:
> Hi,
>
> There seems to be a discrepancy between what the default value of
> 'allocator_may_return_null' should be:
>   source code:
>   ./lib/sanitizer_common/sanitizer_flags.inc:COMMON_FLAG(bool,
> allocator_may_return_null, false,
>
>  docs (https://code.google.com/p/address-sanitizer/wiki/Flags):
> allocator_may_return_null 1 If false, the allocator will crash instead of
> returning 0 on out-of-memory.
>
>
Good catch, fixed the wiki.

> I suspect that the reason to have it default to 'false' is because
that
> would provide for a more predictable out of memory behavior in most use
> cases.
>
Correct.

> Is that the case or should this be system dependent and match the default
> system allocator?
>
Errr.. I don't think so.
asan's allocator is very different from "a system allocator" in
many other
ways.
E.g. it consumes more memory, and so a program that runs normally w/ a
system allocator will OOM with asan.
It's better to have asan crash instantly on OOM rather than returning 0.
The user is then notified about the flag.


>
> Thanks,
> Anna.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150310/ee71064b/attachment.html>