ko1nksm
commented
3 years ago Thank you for investigating.
The CI environment is different from the actual machine, so it may behave strangely in rare cases. I've had a few such experiences. As you know, it is very difficult to resolve it. And travis is no longer supporting the OSS project, so I can't test. I would be applicate if you could help me.
is the debugger like gdb is a choice?
Unfortunately, I don't know how to test using gdb.
shellspec_exists_envkey() {
(
key="$1"
callback() { [ ! "$1" = "$key" ] &&:; }
shellspec_list_envkeys callback && return 1
return 0
) &&:
}This code is very strange and may be resolved by changing it to a more natural code.
I am wondering if subshells are really the cause of the problem. What happens if we don't use return?
# the problem goes away without subshell
shellspec_exists_envkey () { return 0; }
# the problem remains with subshell
shellspec_exists_envkey () { (return 0); }
# ?
shellspec_exists_envkey () { ( : ); }
# ?
shellspec_exists_envkey () { ( exit 0 ); }If all the code using subshells fails, then implement it without subshells.
shellspec_exists_envkey() {
shellspec_exists_envkey_key=$1
shellspec_list_envkeys shellspec_exists_envkey_ && return 1
return 0
}
shellspec_exists_envkey_() { [ ! "$1" = "$shellspec_exists_envkey_key" ]; }I believe this code will work. However, I have not tested it with other shells, so additional modifications may be needed.
alexzhangs
This issue is found only in the environment described below. It works totally well on macOS and AWS Linux 2 AMI (with the same
shellspecandkcovversion and test case).Environment:
spec/foo_spec.sh:
The Segmentation fault is caused by either of the last 2 statements:
variableandresult of function. My debug is focused on thevariablestatement. Make some slight change onxsh.sh, such as removetrapor__xsh_clean, the problem will go away.Command:
Error log:
Since Travis is the only environment the problem is found, the debug is very painful. I inject debug code into
shellspecat the deploy time by.travis.yml.Step by step, finally found the code where the segmentation fault is triggered. Seems the error is triggered by the subshell within the function
shellspec_exists_envkey.I simplified the function code, and I can see the difference, but really don't understand why, don't have a clue what to do next, is the debugger like
gdbis a choice?The following is for reference.
The full .travis.yml:
The system limits on Travis:
The full list of environment variables on Travis: