// |reftest| skip-if(is64Bit||Android) -- No test results /* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ /* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ /* * * Date: 16 July 2002 * SUMMARY: Testing that Array.sort() doesn't crash on very large arrays * See http://bugzilla.mozilla.org/show_bug.cgi?id=157652 * * How large can a JavaScript array be? * ECMA-262 Ed.3 Final, Section 15.4.2.2 : new Array(len) * * This states that |len| must be a a uint32_t (unsigned 32-bit integer). * Note the UBound for uint32's is 2^32 -1 = 0xFFFFFFFF = 4,294,967,295. * * Check: * js> var arr = new Array(0xFFFFFFFF) * js> arr.length * 4294967295 * * js> var arr = new Array(0x100000000) * RangeError: invalid array length * * * We'll try the largest possible array first, then a couple others. * We're just testing that we don't crash on Array.sort(). * * Try to be good about memory by nulling each array variable after it is * used. This will tell the garbage collector the memory is no longer needed. * * As of 2002-08-13, the JS shell runs out of memory no matter what we do, * when trying to sort such large arrays. * * We only want to test that we don't CRASH on the sort. So it will be OK * if we get the JS "out of memory" error. Note this terminates the test * with exit code 3. Therefore we put * * |expectExitCode(3);| * * The only problem will arise if the JS shell ever DOES have enough memory * to do the sort. Then this test will terminate with the normal exit code 0 * and fail. * * Right now, I can't see any other way to do this, because "out of memory" * is not a catchable error: it cannot be trapped with try...catch. * * * FURTHER HEADACHE: Rhino can't seem to handle the largest array: it hangs. * So we skip this case in Rhino. Here is correspondence with Igor Bukanov. * He explains that Rhino isn't actually hanging; it's doing the huge sort: * * Philip Schwartau wrote: * * > Hi, * > * > I'm getting a graceful OOM message on trying to sort certain large * > arrays. But if the array is too big, Rhino simply hangs. Note that ECMA * > allows array lengths to be anything less than Math.pow(2,32), so the * > arrays I'm sorting are legal. * > * > Note below, I'm getting an instantaneous OOM error on arr.sort() for LEN * > = Math.pow(2, 30). So shouldn't I also get one for every LEN between * > that and Math.pow(2, 32)? For some reason, I start to hang with 100% CPU * > as LEN hits, say, Math.pow(2, 31) and higher. SpiderMonkey gives OOM * > messages for all of these. Should I file a bug on this? * * Igor Bukanov wrote: * * This is due to different sorting algorithm Rhino uses when sorting * arrays with length > Integer.MAX_VALUE. If length can fit Java int, * Rhino first copies internal spare array to a temporary buffer, and then * sorts it, otherwise it sorts array directly. In case of very spare * arrays, that Array(big_number) generates, it is rather inefficient and * generates OutOfMemory if length fits int. It may be worth in your case * to optimize sorting to take into account array spareness, but then it * would be a good idea to file a bug about ineficient sorting of spare * arrays both in case of Rhino and SpiderMonkey as SM always uses a * temporary buffer. * */ //----------------------------------------------------------------------------- var BUGNUMBER = 157652; var summary = "Testing that Array.sort() doesn't crash on very large arrays"; var expect = 'No Crash'; var actual = 'No Crash'; printBugNumber(BUGNUMBER); printStatus(summary); expectExitCode(0); expectExitCode(5); try { var a1=Array(0xFFFFFFFF); a1.sort(); a1 = null; var a2 = Array(0x40000000); a2.sort(); a2=null; var a3=Array(0x10000000/4); a3.sort(); a3=null; } catch(ex) { // handle changed 1.9 branch behavior. see bug 422348 expect = 'InternalError: allocation size overflow'; actual = ex + ''; } reportCompare(expect, actual, summary);