patrikaaberg / dmce Goto Github PK

When trying to probe an OpenJDK version I got a lot of error messages. These error messages was received when trying to probe using dmce-launcher -aq on a directory with many files in it (about 4000 files). This issue could be worked around by increasing the limit of allowed open files, using ulimit -n . Here is a snippet of the error messages received:

Exception in thread Thread-6813:
Traceback (most recent call last):
File "/usr/lib/python3.6/threading.py", line 916, in _bootstrap_inner
self.run()
File "/usr/lib/python3.6/threading.py", line 864, in run
self._target(self._args, **self._kwargs)
File "/usr/share/dmce/dmce-copy-headers", line 38, in parse_file
fp = open(f)
OSError: [Errno 24] Too many open files: '/home/samuel/Documents/MasterThesis/openJDKTest/jdk17/test/hotspot/jtreg/vmTestbase/nsk/jvmti/GetArgumentsSize/argsize001/argsize001.cpp'

Exception in thread Thread-6812:
Traceback (most recent call last):
File "/usr/lib/python3.6/threading.py", line 916, in _bootstrap_inner
self.run()
File "/usr/lib/python3.6/threading.py", line 864, in run
self._target(self._args, self._kwargs)
File "/usr/share/dmce/dmce-copy-headers", line 38, in parse_file
fp = open(f)
OSError: [Errno 24] Too many open files: '/home/samuel/Documents/MasterThesis/openJDKTest/jdk17/test/hotspot/jtreg/vmTestbase/nsk/jvmti/SetLocalVariable/setlocal001/setlocal001.cpp'

Exception in thread Thread-6815:
Traceback (most recent call last):
File "/usr/lib/python3.6/threading.py", line 916, in _bootstrap_inner
self.run()
File "/usr/lib/python3.6/threading.py", line 864, in run
self._target(*self._args, self._kwargs)
File "/usr/share/dmce/dmce-copy-headers", line 38, in parse_file
fp = open(f)
OSError: [Errno 24] Too many open files: '/home/samuel/Documents/MasterThesis/openJDKTest/jdk17/test/hotspot/jtreg/vmTestbase/nsk/jvmti/GetArgumentsSize/argsize002/argsize002.cpp'

Currently, when running dmce in a directory that does not contain a git repository, the user is not given any information about the execution:

not/a/repo$ dmce
not/a/repo$

Proposed output:

not/a/repo$ dmce
error: current directory does not contain a git repository
not/a/repo$

I'm unable to insert any probes, following the below steps.

Why is it reporting 'Probes inserted: 0'?

This looks like some very promising work.

Is there a FAQ or support community behind this project?

Thanks,
ksfasf

1. git clone https://github.com/nyuwireless-unipd/ns3-mmwave.git
2. dmce-set-profile -v coverage -i sw/mmwave/examples
3. dmce-launcher -n 2 --progress
   |. . . . . . . . . . . . . . . >|
   |. . . . . . . . . . . . . . . >|
   ===============================================
   git repository /home/dng/ns3-mmwave
   Old SHA-1 0e8842fa (0e8842fa0538a86296532d46dc73059108534b16)
   New SHA-1 HEAD (a402f10160a84d1901145fbcbffc39123ed23ad7)
   Files examined 11
   Files probed 0
   Files skipped 11
   Probes inserted 0
   DMCE version 1.7-1
   ===============================================
   done: real: 0m9.113s user: 0m27.398s sys: 0m1.470s

Regarding this execution, the time taken for the command to return control of the interactive shell was around 2 minutes for 32 probes.

Can this elapsed time be reduced?

I have been using DMCE on our code base for running tests daily. We needed to collect historical data so besides running tests daily, I am also running a script that iterates through our git history and runs the tests in an iterative manner on a day-2-day basis.

DMCE runs fine most of the times but we have a scenario where the number of unique probes traversed was more than the actual number of probes resulting into coverage % of 131.4.

Coverage %: 131.4 %
Total Number of Probes: 35
Unique probes: 46
Revisited Probes: 77

Maybe there is a corner case. More details and logs can be provided on request.

The regex expression used in print_summary.py for checking whether a probe is executed or not can also match output errors from the compiler. A proper regex should check if a probe number is printed instead.

i.e.

.DMCE_PROBE((.)).* should be .DMCE_PROBE((\d+)).

I am providing a PR for the same.

Is there a support forum or similar for dmce?

Do you have any usage statistics? How big is the user base?

OpenJDK 11u, 2c47c45f0747e86a3b97384bea48107627ffd040:

Instrumenting the following line 102 in src/hotspot/share/oops/objArrayKlass.cpp:

  if (!element_klass->is_instance_klass() ||
      (name = InstanceKlass::cast(element_klass)->array_name()) == NULL) {

leads to the semantically different code below:

  if (!(DMCE_PROBE(25201), element_klass->is_instance_klass() ||
      (name = InstanceKlass::cast(element_klass)->array_name()) == NULL)) {

Ignoring the DMCE_PROBE(), the code above replaces (!a || b) by (!(a || b)), i.e., it negates the entire disjunction rather than just the left-hand side of the ||.

A semantics-preserving replacement would be:

  if (DMCE_PROBE(25201), !element_klass->is_instance_klass() ||
    (name = InstanceKlass::cast(element_klass)->array_name()) == NULL) {

which works as intended.

dmce-configure-local sets
DMCE_POST_HOOK:

When changing profile to racecheck with dmce-set-profile -p -d /PATH/TO/DMCE racecheck the code in setProbeOrProlog misfires and sets DMCE_POST_HOOK:/dmce-post-hook-racheck, which is not a valid path.