On 7 August 2018 at 13:22, Ben Coman <btc@openinworld.com> wrote:

Next is to try��loading an Image.

+ Downloaded and unzipped�� http://files.pharo.org/image/70/latest-minimal-64.zip

+ In Visual Studio, right-clicked CMakeLists.txt > Debug and Launch Settings > pharo.exe

+ Edited the presented "launch.vs.json" as follows...

{

�� "version": "0.2.1",

�� "defaults": {},

�� "configurations": [

�� {

�� "type": "default",

�� "project": "CMakeLists.txt",

�� "projectTarget": "pharo.exe (dist\\pharo.exe)",

�� "name": "pharo.exe with image (dist\\pharo.exe)",

�� "args": [

�� "C:\\#Dev\\latest-minimal-64\\Pharo7.0-metacello-64bit-65cff7b.image",

�� "s"

�� ]

�� },

�� ]

}

Now running "x64-Debug" and��"pharo.exe with image"

I find it gets hung up in...

�� sqAllocateMemory(usqInt minHeapSize, usqInt desiredHeapSize)

in call...

�� alloc = sqAllocateMemorySegmentOfSizeAboveAllocatedSizeInto (roundUpToPage(desiredHeapSize), address, &allocBytes);

with...

�� desiredHeapSize ==> 39583744 (~39MB)

�� address ==> 0x00007ff7d6e00000

where inside that call...��

�� void *sqAllocateMemorySegmentOfSizeAboveAllocatedSizeInto(sqInt size, void *minAddress, sqInt *allocatedSizePointer)"

has...��

�� size ==> 39583744

�� minAddress ==> 0x00007ff7d6e00000

�� address ==> 0x00000000d6e00000

�� bytes ==> 39583744 (~39MB)

�� delta ==> 1048576

it gets stuck the following loop...�� https://github.com/bencoman/opensmalltalk-vm/blob/8231b96f9b/platforms/win32/vm/sqWin32SpurAlloc.c#L139-L183

�� while ((usqIntptr_t)(address + bytes) > (usqIntptr_t)address)

�� {�� alloc = VirtualAlloc(address, bytes, MEM_COMMIT|MEM_RESERVE, PAGE_READWRITE);

�� ...

�� if (alloc >= (char *)minAddress) && (alloc <= address + delta))�� return alloc;

�� ...

�� address += delta;

�� }

where...��

�� alloc ==> 0x00000000d6e00000

�� (alloc >= (char *)minAddress) ==> false.

which is effectively...�� (does this seem normal?)

�� ((char *)minAddress-address)/delta ==> 134,180,864 loops before exit.

but...��

�� (alloc >= (char *)address) ==> true.

would exit on the first pass of the loop, except...��

�� /* For some reason (large page support?) we can ask for a page-aligned

�� * address such as 0xNNNNf000 but VirtualAlloc will answer 0xNNNN0000.

�� * So accept allocs above minAddress rather than allocs above address

�� */

where back one level "minAddress" is derived like this...

�� /* choose a suitable starting point. In MinGW the malloc heap is below the

�� * program, so take the max of a malloc and something form uninitialized

�� * data.

�� */

hint = malloc(1);

free(hint);

hint = max(hint,(char *)&fIsConsole);

�� address = (char *)(((usqInt)hint + alignment - 1) & ~(alignment - 1));

�� address passed-by-value into minAddress��

Now I'm out of my depth trying to understand the background on that logic,

but if I blindly change the loop exit condition��

from...

�� (alloc >= (char *)minAddress)

to...

�� (alloc >= (char *)address)

then�� osvm_loadImage(tempImageNameAttempt()

and subsequently�� osvm_loadDefaultImage()�� succeed.��

So some expert attention would be good here (as time permits)��

In the meantime, I'll forge ahead to play with�� osvm_run();

cheers -ben

P.S. making "#define printProbes 1"�� and "fIsConsole = 1;"

shows...

probing [00000000D6E00000,00000000D93C0000)

probing [00000000D6F00000,00000000D94C0000)

probing [00000000D7000000,00000000D95C0000)

probing [00000000D7100000,00000000D96C0000)

probing [00000000D7200000,00000000D97C0000)

probing [00000000D7300000,00000000D98C0000)

probing [00000000D7400000,00000000D99C0000)

probing [00000000D7500000,00000000D9AC0000)

probing [00000000D7600000,00000000D9BC0000)