[Webkit-unassigned] [Bug 144677] New stress/new-largeish-contiguous-array-with-size.js fails on 32 bit platforms

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=144677

--- Comment #2 from Filip Pizlo <fpizlo at apple.com> ---
(In reply to comment #0)
> New stress/new-largeish-contiguous-array-with-size.js
> introduced in https://trac.webkit.org/changeset/183847
> and fails on 32 bit platforms.
> 
> It failed once on GTK 32 bit bot:
> -
> https://build.webkit.org/builders/GTK%20Linux%2032-bit%20Release/builds/52461
> - stress/new-largeish-contiguous-array-with-size.js.default: Exception:
> Error: heap too big before forced GC: 108104384
> 
> It fails persistently on EFL 32 bit ARM bots since it is introduced:
> - stress/new-largeish-contiguous-array-with-size.js.default: Exception:
> Error: heap too big before forced GC: 119708840
> - stress/new-largeish-contiguous-array-with-size.js.default: Exception:
> Error: heap too big before forced GC: 124108928
> - stress/new-largeish-contiguous-array-with-size.js.default: Exception:
> Error: heap too big before forced GC: 129308808
> - stress/new-largeish-contiguous-array-with-size.js.default: Exception:
> Error: heap too big before forced GC: 120108656
> ...
> 
> Can it be a real bug or a bug in the test?

I think that the test might be wrong.  I was assuming that 100MB is the largest that our minimum heap size would get.  But I'm wrong:

static inline size_t minHeapSize(HeapType heapType, size_t ramSize)
{
    if (heapType == LargeHeap)
        return min(largeHeapSize, ramSize / 4);
    return smallHeapSize;
}

So, if you're on a machine with tons of heap, we'll be happy to give you 1/4 RAM headroom.  This actually seems super sketchy anyways and maybe we should fix that - but for now it means that as written, my test is incorrect.

I'll investigate.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20150507/36fa42c4/attachment-0001.html>