Thanks for the follow up. Glad to hear its nothing intrinsic to Pharo. btw, Is there somewhere in the documentation or class/method comments that you believe this N-base situation should be spelled out better? cheers -ben On Wed, May 3, 2017 at 8:32 PM, Raffaello Giulietti < raffaello.giulietti@lifeware.ch> wrote:
Problem solved.
The cause was a mismatch between a 0-based index and a 1-based index on copying arrays between Smalltalk and C.
Rather confusingly, for some reason, FFI indices are 1-based even for C arrays, so that calling ExternalAddress>>byteAt:put:, for example, requires the index to be unnaturally 1-based.
While I put a lot of care at this issue, a specific spot in the code was erroneous, leading to an improperly terminated C-style string which, in turn, caused the indirect crash of Pharo.
My fault, however, neither Pharo's nor the JVM's.
Thanks to all involved in replies.
Greetings Raffaello
On 2017-05-03 13:52, Ben Coman wrote:
On Wed, May 3, 2017 at 4:33 PM, Raffaello Giulietti < raffaello.giulietti@lifeware.ch <mailto:raffaello.giulietti@lifeware.ch>> wrote:
Unfortunately, as I just discovered, older versions of Pharo do not support UFFI. I don't have the time (nor a real interest) in backporting the code to an older pre-Spur VM which has probably reached its EOL just for the sake of experimentation. It would probably take hours to understand the older FFI model and to adapt the code for no real advantage.
The current Pharo VM seems to support only the --memory option: I tried with 1 and 2 GiB, nothing changes in the misbehavior.
I also tried the two heap-related JVM options in isolation, unsuccessfully. Just for information, the current values for the options are -Xms16m and -Xmx512m, meaning that the JVM heap should start with 16 MiB and grow to a maximum size of 512 MiB, both of which are rather modest values even for a 32 bit process. Besides, they work in VisualWorks (32 bit).
What about tiny memory allocation like --memory 100m and -Xmx 100m ?
Can you share some code that causes the crash?
cheers -ben
The StackOverflow page indicated below is from year 2010, during the age of Java 6. The JVM has undergone major changes since then, including memory management, so I'm not sure the discussion there is relevant for Java 8, the release we are targeting as of today. But even if it were, I still cannot understand the crash in Pharo.
On 2017-05-03 06:36, Ben Coman wrote:
On Wed, May 3, 2017 at 12:30 AM, Denis Kudriashov <dionisiydk@gmail.com <mailto:dionisiydk@gmail.com> <mailto:dionisiydk@gmail.com <mailto:dionisiydk@gmail.com>>> wrote:
Could you check how it works with Pharo5 which was based on prespur CogVM?
Note, Pharo 5 Release was a Spur VM. Try using... http://files.pharo.org/get-files/50-preSpur/ <http://files.pharo.org/get-files/50-preSpur/> http://files.pharo.org/image/50-preSpur/50495.zip <http://files.pharo.org/image/50-preSpur/50495.zip>
or even Pharo 4.
cheers -ben
2017-05-02 16:36 GMT+02:00 Raffaello Giulietti <raffaello.giulietti@lifeware.ch <mailto:raffaello.giulietti@lifeware.ch> <mailto:raffaello.giulietti@lifeware.ch
<mailto:raffaello.giulietti@lifeware.ch>>>:
Hello,
I'm on Pharo 6 (32 bit)/Windows 10 (64 bit).
I'm trying to load and use the (32 bit) JVM DLL into a running Pharo image via the UFFI.
Everything works correctly, including JVM method invocations, except when trying to set the minimum and maximum JVM heap size by passing the -Xms<size> and -Xmx<size> options upon JVM creation. With these options, Pharo simply crashes without leaving any trace.
Can you isolate whether the problem is either one of those, or it happens with both, and if there is some cutoff where it works? Also, perhaps try limiting heap being used by Pharo --memory <size>[mk] use fixed heap size (added to image size) --mmap <size>[mk] limit dynamic heap size (default: 1024m) [out 2G on 32-bit) --maxoldspace <size>[mk] set max size of old space memory to bytes (disclaimer, I'm not too knowledgeable on these, just pulled them from --help)
Also consider that heap needs to be contiguous... **http://stackoverflow.com/questions/2457514/understanding- max-jvm-heap-size-32bit-vs-64bit <http://stackoverflow.com/questions/2457514/understanding- max-jvm-heap-size-32bit-vs-64bit> cheers -ben
Apparently, memory management requests from the JVM interfere with Pharo's own memory management.
Please note that we successfully use the same mechanism for both VisualWorks (32 bit)/Windows 10 (64 bit) and Gemstone 64 bit/Linux, so it seems it has to do with a Pharo limitation somehow.
Further, I couldn't find any documentation on how to increase Pharo's working set.
So the questions are: * What is the most probable cause of the crash described above? * Where is there more doc about Pharo's memory configuration settings?
Greetings Raffaello