Fix bugs in argument passing to scripts#177
Conversation
|
Sorry, was trying out Copilot to see if it could find issues that I couldn't, and it posted without my permission! Not trying that again... |
|
Both those are valid finds, and good ones. The extra readlink is debugging code I accidently left in and the stat one is accurate. I'll fix those both. It got some details wrong. The readlink not infinite recursion, just a performance hit, but still valid. |
|
Regardless, it does look like Copilot found what I think may be a real problem, though: the wrapper for |
| debug_printf2("Exec search request returned %s -> %s\n", newexec, *reloc_exec ? *reloc_exec : "NULL"); | ||
| if (*reloc_exec) { | ||
| if (orig_file_abspath) { | ||
| *orig_file_abspath = strdup(newexec); |
There was a problem hiding this comment.
A question: the added code above in adjust_if_script uses spindle_strdup() (and replaces a strdup() with spindle_strdup(), but here strdup() is used. Under what circumstances should one be used vs the other?
There was a problem hiding this comment.
I'll move it to spindle_strdup() for consistency.
At one point I had a memory heap corruption checker integrated with the spindle, which required all of the client heap calls to go through wrappers (I had to do the interception manually since the library namespace issues kept spindle from working with regular tools).
| ./run_driver --serial ./interpreter_test | ||
| CHECK_RETCODE | ||
|
|
||
| ./run_driver --serial ./exec_shell |
There was a problem hiding this comment.
Unlike the other tests, this one doesn't print "PASSED." when it passes. For consistency, we might want to have it do that, so that it doesn't look like it crashed without printing.
Running: ./run_driver --serial ./interpreter_test
PASSED.
Running: ./run_driver --serial ./exec_shell
ALL TESTS PASSED
|
I addressed the identified issues. |
2 participants
We had some failures in an application that came down to Spindle changing how the $1 in a bash script changed depending on whether it was passed in or out of spindle. This is related to #96, which was essentially the same issue, though this extends the fix into more corner cases. Specifically when argv[0] of the script does not match the argument passed to the first parameter of exec.
In addition, this adds a more comprehensive test case that tries to exec a set of scripts under a combinatorial set of circumstances (exec vs execve, argv[0] matches exec or not, path search vs relative vs absolute, and shell type).