harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Egor Pasko <egor.pa...@gmail.com>
Subject Re: [doc] are KNOWN ISSUES and TODO items valid?
Date Tue, 12 Dec 2006 13:55:54 GMT
On the 0x23C day of Apache Harmony Alexey Varlamov wrote:
> 11 Dec 2006 21:20:23 +0600, Egor Pasko <egor.pasko@gmail.com>:
> > On the 0x23A day of Apache Harmony Alexey Varlamov wrote:
> > > I've looked through "Known issues" page and dropped few items, which
> > > were resolved already:
> > > - Java assertions support
> > > - StackOverflowError support for JIT mode
> >
> > why?
> > we do not have SOE 100% safe yet, JIT algorithms can be caught with
> > SOE on large bytecodes. We should estimate how much stack we can take
> > before each graph traversal activity. Not done yet and leads to a crash.
> >
> The difference is, a technical imperfection vs missing feature (which
> was originally listed) - so feel free to add a detailed TODO record
> instead.


I made a JIT task from that under the new chapter "JIT Stability":

4. Avoiding stack overflow in JIT C++ code

Problem: If compiled method's size is large or in case of aggressive
inlining parameters are set in Jitrino.OPT the latter can crash with
stack overflow while recursively traversing the CFG (=Control Flow

Task: To avoid stack overflows, estimate the amount of space allocated
by each optimization stage, check that it does not exceed the
available free space on the stack. VM provides necessary functionality
to check that via VM helper function get_available_stack_size(). In
dangerous situations stop inlining or return JIT_FAILURE so that other
JITs (for example, Jitrino.JET) can pick the compilation task.


if there is a JIRA issue for that, feel free to add a reference.

Egor Pasko

View raw message