Memory leak in 'setTimeout' on exceptions

[László Langó]

IoT.js version:

Checked revision: 67b1de3
Build command: tools/build.py --clean

OS:

Ubuntu 16.04

Test case:

setTimeout(function() { throw "error" }, 100);

Backtrace:

The test case a memory leak.

with x86_64-linux target:

==30636== Memcheck, a memory error detector
==30636== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al.
==30636== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info
==30636== Command: ./build/x86_64-linux/debug/bin/iotjs tmp.js
==30636== 
uncaughtException: error
==30636== 
==30636== HEAP SUMMARY:
==30636==     in use at exit: 104 bytes in 1 blocks
==30636==   total heap usage: 41 allocs, 40 frees, 5,257 bytes allocated
==30636== 
==30636== 104 bytes in 1 blocks are definitely lost in loss record 1 of 1
==30636==    at 0x4C2FB55: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==30636==    by 0x41608F: iotjs_buffer_allocate (iotjs_util.c:68)
==30636==    by 0x42586D: iotjs_timerwrap_create (iotjs_module_timer.c:27)
==30636==    by 0x4262FB: Timer (iotjs_module_timer.c:156)
==30636==    by 0x4627BF: ecma_op_function_call (ecma-function-object.c:467)
==30636==    by 0x462BF1: ecma_op_function_construct_simple_or_external (ecma-function-object.c:598)
==30636==    by 0x462DB6: ecma_op_function_construct (ecma-function-object.c:666)
==30636==    by 0x47A7F1: opfunc_construct (vm.c:483)
==30636==    by 0x47F679: vm_execute (vm.c:2798)
==30636==    by 0x47F8BF: vm_run (vm.c:2873)
==30636==    by 0x462747: ecma_op_function_call (ecma-function-object.c:406)
==30636==    by 0x47A6B3: opfunc_call (vm.c:425)
==30636== 
==30636== LEAK SUMMARY:
==30636==    definitely lost: 104 bytes in 1 blocks
==30636==    indirectly lost: 0 bytes in 0 blocks
==30636==      possibly lost: 0 bytes in 0 blocks
==30636==    still reachable: 0 bytes in 0 blocks
==30636==         suppressed: 0 bytes in 0 blocks
==30636== 
==30636== For counts of detected and suppressed errors, rerun with: -v
==30636== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)

[László Langó]

Status update: I have a fix for this, but I cannot add the testcase because the lack of the JS testrunner. I must fix the testrunner at first.

[Sanggyu Lee]

@LaszloLango Could you explain more about the lack of the JS testrunner? In fact, our plan is to replace JS testrunner with python testrunner. So I would not put more time on JS test runner.

Another question occurred: Is there any reason we cannot replace JS testrunner with python runner in travis?

[László Langó]

@glistening the problem is that if a test fails (even it is an expected fail) the environment does not clean up properly. For example active timers could run in the background and can trigger false negative test failures later. AFAIK we can replace the JS testrunner with the python testrunner. It could make things easier. I'll open a PR for that. Should we keep the JS testrunner in the repository or remove it completely?

[Sanggyu Lee]

@LaszloLango Please replace JS testrunner, but we need time to inform the potential JS test user that it will be removed. What about adding message for JS testrunner invocation like "JS testrunner will be removed, please use python runner instead.", and keep JS testrunner for some days, and remove it?

[László Langó]

@glistening Ok, we can do it.

[László Langó]

@glistening I had some issue when tried to replace the testrunner, so I am not sure when I could upload the final patch.

[Sanggyu Lee]

@LaszloLango Could you share the issue here or by IRC? We would like to deprecate JS testrunner to reduce maintenance cost.

[László Langó]

@glistening sorry for the late answer. One of my biggest problems at the moment is the mixed usage of travis_check.py and precommit.py. The second problem is to build and use the 32bit binary on 64bit systems (needed for address sanitizer job).

[Sanggyu Lee]

@LaszloLango I can't understand fully what is related to deprecating JS test runner and the mixed usage of travis_check.py and precommit.py. Could you explain more in #1229? If merging into one file helps deprecating JS testrunner, it could be a good backing example.

Second, do you mean python runner could not run 32bit binary on 64 bit systems? We may add the feature js-test-runner after completing js-test-runner restructuring.

[László Langó]

@glistening not exactly. The testrunner is able to run any binary. The problem comes from the separation of the build and the test. I wanted to change the travis jobs to use the python testrunner. This is why travis_check.py and precommit.py related. I'll open a PR soon.