Date: Tue, 06 Jun 2000 23:27:16 -0700
From: "Steffen Grarup" <###@###.###>
Subject: Re: request for comment
To: "jon.masamitsu" <###@###.###>
Cc: ###@###.###

Jon,

Please make that change. It should be completely safe and is certainly
better. You may want to combine the two:

> < if (!succeeded) {
> ---
> > if (!succeeded && free() < size*wordSize) {

That would have slightly better performance (probably in the noise, though).

/Steffen




"jon.masamitsu" wrote:
>
> Currently,
>
> oop* OneSpaceOldGeneration::safepoint_allocate_and_expand(int size)
>
> does a hard memory error exit - calls
>
> vm_exit_out_of_memory(size*wordSize, "heap expansion");^M
>
> if the heap cannot be expanded by the requested "size". Even though
> the expand may not have expanded by "size", it may have expanded
> enough to allow the allocation - the expansion may go to the limit
> of the heap and that plus the free space previously in the heap may
> be enough. I'd like to change the test as below. Instead of using
> the return value from the call to expand(), check the free space
> before calling for a hard memory error exit.
>
> Comments?
>
> PS. The current code can lead to the start of a scavenge that should
> succeed (there is enough old generation space to guarantee the
> worst case evacuation) but fails due to the failure to expand
> the full "size".
>
> PPS. I expect to be at JavaOne tomorrow and will gather up the comments
> when I get back to the office.
>
> ------- generation.cpp -------
> 480c481,482
> < if (!succeeded) {
> ---
> > // expand() may not have fully succeeded but still expanded enough.
> > if (free() < size*wordSize) {