Honggyu Kim
Honggyu Kim
There are a lot of system stat information in guider, but it's difficult to understand what each field means. I think it'd be better to have a document that explains...
The test `t048_malloc_impl.py` gets `TM` in aarch64 as follows. ``` $ ./runtest.py 048 Start 1 tests with 1 worker Test case pg finstrument-fu ------------------------: O0 O1 O2 O3 Os O0...
It can be helpful to have an option that tests if some features are available to use in the currently running system. For example, if `/proc/sys/kernel/perf_event_paranoid` is `3`, then uftrace...
It seems very weird that some tests are randomly failed. For example, 226 test result is different for each execution. ``` $ ./runtest.py 226 Start 1 tests with 1 worker...
The current `--graphviz` output for `t-abc` is as follows. ```dot $ uftrace record t-abc $ uftrace dump --graphviz ... digraph "t-abc" { "__monstartup" [xlabel = "1"] "__cxa_atexit" [xlabel = "1"]...
`uftrace graph` actually doesn't show a call graph, but just a call tree output. For example, `t-fibonacci` example's output is as follows. ``` $ gcc -pg -o t-fibonacci tests/s-fibonacci.c $...
uftrace currently support chrome trace viewer format with `uftrace dump --chrome` and that is really useful to visualize the workload. However, there are some requests to support firefox profiler format...
We don't have to lint everything in our python script, but it'd be better to follow common rules in our python code. The following warnings from [pylint](https://github.com/PyCQA/pylint) are better to...
In some cases, debug info can be in separate files. Since the current uftrace implementation doesn't support such cases, it can fail to read symbol tables and debug info. Here...
It seems that the test speed is a lot faster with multiprocessing pool, but it makes the result unreliable sometimes. I've seen many random failures as I posted it in...